2 * Copyright (c) 2009, 2010, 2011, 2012 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.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
208 static uint64_t pick_datapath_id(const struct ofproto *);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto *);
211 static void update_mtu(struct ofproto *, struct ofport *);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class **ofproto_classes;
218 static size_t n_ofproto_classes;
219 static size_t allocated_ofproto_classes;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
224 /* Initial mappings of port to OpenFlow number mappings. */
225 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
229 /* Must be called to initialize the ofproto library.
231 * The caller may pass in 'iface_hints', which contains an shash of
232 * "iface_hint" elements indexed by the interface's name. The provider
233 * may use these hints to describe the startup configuration in order to
234 * reinitialize its state. The caller owns the provided data, so a
235 * provider will make copies of anything required. An ofproto provider
236 * will remove any existing state that is not described by the hint, and
237 * may choose to remove it all. */
239 ofproto_init(const struct shash *iface_hints)
241 struct shash_node *node;
244 ofproto_class_register(&ofproto_dpif_class);
246 /* Make a local copy, since we don't own 'iface_hints' elements. */
247 SHASH_FOR_EACH(node, iface_hints) {
248 const struct iface_hint *orig_hint = node->data;
249 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
250 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
252 new_hint->br_name = xstrdup(orig_hint->br_name);
253 new_hint->br_type = xstrdup(br_type);
254 new_hint->ofp_port = orig_hint->ofp_port;
256 shash_add(&init_ofp_ports, node->name, new_hint);
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->init(&init_ofp_ports);
264 /* 'type' should be a normalized datapath type, as returned by
265 * ofproto_normalize_type(). Returns the corresponding ofproto_class
266 * structure, or a null pointer if there is none registered for 'type'. */
267 static const struct ofproto_class *
268 ofproto_class_find__(const char *type)
272 for (i = 0; i < n_ofproto_classes; i++) {
273 const struct ofproto_class *class = ofproto_classes[i];
278 class->enumerate_types(&types);
279 found = sset_contains(&types, type);
280 sset_destroy(&types);
286 VLOG_WARN("unknown datapath type %s", type);
290 /* Registers a new ofproto class. After successful registration, new ofprotos
291 * of that type can be created using ofproto_create(). */
293 ofproto_class_register(const struct ofproto_class *new_class)
297 for (i = 0; i < n_ofproto_classes; i++) {
298 if (ofproto_classes[i] == new_class) {
303 if (n_ofproto_classes >= allocated_ofproto_classes) {
304 ofproto_classes = x2nrealloc(ofproto_classes,
305 &allocated_ofproto_classes,
306 sizeof *ofproto_classes);
308 ofproto_classes[n_ofproto_classes++] = new_class;
312 /* Unregisters a datapath provider. 'type' must have been previously
313 * registered and not currently be in use by any ofprotos. After
314 * unregistration new datapaths of that type cannot be opened using
315 * ofproto_create(). */
317 ofproto_class_unregister(const struct ofproto_class *class)
321 for (i = 0; i < n_ofproto_classes; i++) {
322 if (ofproto_classes[i] == class) {
323 for (i++; i < n_ofproto_classes; i++) {
324 ofproto_classes[i - 1] = ofproto_classes[i];
330 VLOG_WARN("attempted to unregister an ofproto class that is not "
335 /* Clears 'types' and enumerates all registered ofproto types into it. The
336 * caller must first initialize the sset. */
338 ofproto_enumerate_types(struct sset *types)
342 for (i = 0; i < n_ofproto_classes; i++) {
343 ofproto_classes[i]->enumerate_types(types);
347 /* Returns the fully spelled out name for the given ofproto 'type'.
349 * Normalized type string can be compared with strcmp(). Unnormalized type
350 * string might be the same even if they have different spellings. */
352 ofproto_normalize_type(const char *type)
354 return type && type[0] ? type : "system";
357 /* Clears 'names' and enumerates the names of all known created ofprotos with
358 * the given 'type'. The caller must first initialize the sset. Returns 0 if
359 * successful, otherwise a positive errno value.
361 * Some kinds of datapaths might not be practically enumerable. This is not
362 * considered an error. */
364 ofproto_enumerate_names(const char *type, struct sset *names)
366 const struct ofproto_class *class = ofproto_class_find__(type);
367 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
371 ofproto_create(const char *datapath_name, const char *datapath_type,
372 struct ofproto **ofprotop)
374 const struct ofproto_class *class;
375 struct ofproto *ofproto;
381 ofproto_unixctl_init();
383 datapath_type = ofproto_normalize_type(datapath_type);
384 class = ofproto_class_find__(datapath_type);
386 VLOG_WARN("could not create datapath %s of unknown type %s",
387 datapath_name, datapath_type);
391 ofproto = class->alloc();
393 VLOG_ERR("failed to allocate datapath %s of type %s",
394 datapath_name, datapath_type);
399 memset(ofproto, 0, sizeof *ofproto);
400 ofproto->ofproto_class = class;
401 ofproto->name = xstrdup(datapath_name);
402 ofproto->type = xstrdup(datapath_type);
403 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
404 hash_string(ofproto->name, 0));
405 ofproto->datapath_id = 0;
406 ofproto_set_flow_eviction_threshold(ofproto,
407 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
408 ofproto->forward_bpdu = false;
409 ofproto->fallback_dpid = pick_fallback_dpid();
410 ofproto->mfr_desc = NULL;
411 ofproto->hw_desc = NULL;
412 ofproto->sw_desc = NULL;
413 ofproto->serial_desc = NULL;
414 ofproto->dp_desc = NULL;
415 ofproto->frag_handling = OFPC_FRAG_NORMAL;
416 hmap_init(&ofproto->ports);
417 shash_init(&ofproto->port_by_name);
418 simap_init(&ofproto->ofp_requests);
419 ofproto->max_ports = OFPP_MAX;
420 ofproto->tables = NULL;
421 ofproto->n_tables = 0;
422 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
423 ofproto->state = S_OPENFLOW;
424 list_init(&ofproto->pending);
425 ofproto->n_pending = 0;
426 hmap_init(&ofproto->deletions);
427 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
428 ofproto->first_op = ofproto->last_op = LLONG_MIN;
429 ofproto->next_op_report = LLONG_MAX;
430 ofproto->op_backoff = LLONG_MIN;
431 ofproto->vlan_bitmap = NULL;
432 ofproto->vlans_changed = false;
433 ofproto->min_mtu = INT_MAX;
435 error = ofproto->ofproto_class->construct(ofproto);
437 VLOG_ERR("failed to open datapath %s: %s",
438 datapath_name, strerror(error));
439 ofproto_destroy__(ofproto);
443 /* The "max_ports" member should have been set by ->construct(ofproto).
444 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
445 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
446 bitmap_set1(ofproto->ofp_port_ids, 0);
448 /* Check that hidden tables, if any, are at the end. */
449 assert(ofproto->n_tables);
450 for (i = 0; i + 1 < ofproto->n_tables; i++) {
451 enum oftable_flags flags = ofproto->tables[i].flags;
452 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
454 assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
457 ofproto->datapath_id = pick_datapath_id(ofproto);
464 /* Must be called (only) by an ofproto implementation in its constructor
465 * function. See the large comment on 'construct' in struct ofproto_class for
468 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
470 struct oftable *table;
472 assert(!ofproto->n_tables);
473 assert(n_tables >= 1 && n_tables <= 255);
475 ofproto->n_tables = n_tables;
476 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
477 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
482 /* To be optionally called (only) by an ofproto implementation in its
483 * constructor function. See the large comment on 'construct' in struct
484 * ofproto_class for details.
486 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
487 * will then ensure that actions passed into the ofproto implementation will
488 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
489 * function is not called, there will be no such restriction.
491 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
492 * the 'max_ports' restriction. */
494 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
496 assert(max_ports <= OFPP_MAX);
497 ofproto->max_ports = max_ports;
501 ofproto_get_datapath_id(const struct ofproto *ofproto)
503 return ofproto->datapath_id;
507 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
509 uint64_t old_dpid = p->datapath_id;
510 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
511 if (p->datapath_id != old_dpid) {
512 /* Force all active connections to reconnect, since there is no way to
513 * notify a controller that the datapath ID has changed. */
514 ofproto_reconnect_controllers(p);
519 ofproto_set_controllers(struct ofproto *p,
520 const struct ofproto_controller *controllers,
521 size_t n_controllers, uint32_t allowed_versions)
523 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
528 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
530 connmgr_set_fail_mode(p->connmgr, fail_mode);
533 /* Drops the connections between 'ofproto' and all of its controllers, forcing
534 * them to reconnect. */
536 ofproto_reconnect_controllers(struct ofproto *ofproto)
538 connmgr_reconnect(ofproto->connmgr);
541 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
542 * in-band control should guarantee access, in the same way that in-band
543 * control guarantees access to OpenFlow controllers. */
545 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
546 const struct sockaddr_in *extras, size_t n)
548 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
551 /* Sets the OpenFlow queue used by flows set up by in-band control on
552 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
553 * flows will use the default queue. */
555 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
557 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
560 /* Sets the number of flows at which eviction from the kernel flow table
563 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
565 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
566 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
568 ofproto->flow_eviction_threshold = threshold;
572 /* If forward_bpdu is true, the NORMAL action will forward frames with
573 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
574 * the NORMAL action will drop these frames. */
576 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
578 bool old_val = ofproto->forward_bpdu;
579 ofproto->forward_bpdu = forward_bpdu;
580 if (old_val != ofproto->forward_bpdu) {
581 if (ofproto->ofproto_class->forward_bpdu_changed) {
582 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
587 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
588 * 'idle_time', in seconds, and the maximum number of MAC table entries to
591 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
594 if (ofproto->ofproto_class->set_mac_table_config) {
595 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
601 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
603 return connmgr_set_snoops(ofproto->connmgr, snoops);
607 ofproto_set_netflow(struct ofproto *ofproto,
608 const struct netflow_options *nf_options)
610 if (nf_options && sset_is_empty(&nf_options->collectors)) {
614 if (ofproto->ofproto_class->set_netflow) {
615 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
617 return nf_options ? EOPNOTSUPP : 0;
622 ofproto_set_sflow(struct ofproto *ofproto,
623 const struct ofproto_sflow_options *oso)
625 if (oso && sset_is_empty(&oso->targets)) {
629 if (ofproto->ofproto_class->set_sflow) {
630 return ofproto->ofproto_class->set_sflow(ofproto, oso);
632 return oso ? EOPNOTSUPP : 0;
636 /* Spanning Tree Protocol (STP) configuration. */
638 /* Configures STP on 'ofproto' using the settings defined in 's'. If
639 * 's' is NULL, disables STP.
641 * Returns 0 if successful, otherwise a positive errno value. */
643 ofproto_set_stp(struct ofproto *ofproto,
644 const struct ofproto_stp_settings *s)
646 return (ofproto->ofproto_class->set_stp
647 ? ofproto->ofproto_class->set_stp(ofproto, s)
651 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
652 * 'enabled' member of 's' is false, then the other members are not
655 * Returns 0 if successful, otherwise a positive errno value. */
657 ofproto_get_stp_status(struct ofproto *ofproto,
658 struct ofproto_stp_status *s)
660 return (ofproto->ofproto_class->get_stp_status
661 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
665 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
666 * in 's'. The caller is responsible for assigning STP port numbers
667 * (using the 'port_num' member in the range of 1 through 255, inclusive)
668 * and ensuring there are no duplicates. If the 's' is NULL, then STP
669 * is disabled on the port.
671 * Returns 0 if successful, otherwise a positive errno value.*/
673 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
674 const struct ofproto_port_stp_settings *s)
676 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
678 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
679 ofproto->name, ofp_port);
683 return (ofproto->ofproto_class->set_stp_port
684 ? ofproto->ofproto_class->set_stp_port(ofport, s)
688 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
689 * 's'. If the 'enabled' member in 's' is false, then the other members
690 * are not meaningful.
692 * Returns 0 if successful, otherwise a positive errno value.*/
694 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
695 struct ofproto_port_stp_status *s)
697 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
699 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
700 "port %"PRIu16, ofproto->name, ofp_port);
704 return (ofproto->ofproto_class->get_stp_port_status
705 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
709 /* Queue DSCP configuration. */
711 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
712 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
713 * to implement QoS. Instead, it is used to implement features which require
714 * knowledge of what queues exist on a port, and some basic information about
717 * Returns 0 if successful, otherwise a positive errno value. */
719 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
720 const struct ofproto_port_queue *queues,
723 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
726 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
727 ofproto->name, ofp_port);
731 return (ofproto->ofproto_class->set_queues
732 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
736 /* Connectivity Fault Management configuration. */
738 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
740 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
742 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
743 if (ofport && ofproto->ofproto_class->set_cfm) {
744 ofproto->ofproto_class->set_cfm(ofport, NULL);
748 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
749 * basic configuration from the configuration members in 'cfm', and the remote
750 * maintenance point ID from remote_mpid. Ignores the statistics members of
753 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
755 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
756 const struct cfm_settings *s)
758 struct ofport *ofport;
761 ofport = ofproto_get_port(ofproto, ofp_port);
763 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
764 ofproto->name, ofp_port);
768 /* XXX: For configuration simplicity, we only support one remote_mpid
769 * outside of the CFM module. It's not clear if this is the correct long
770 * term solution or not. */
771 error = (ofproto->ofproto_class->set_cfm
772 ? ofproto->ofproto_class->set_cfm(ofport, s)
775 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
776 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
781 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
782 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
783 * 0 if LACP partner information is not current (generally indicating a
784 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
786 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
788 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
789 return (ofport && ofproto->ofproto_class->port_is_lacp_current
790 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
796 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
797 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
798 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
799 * configuration plus, if there is more than one slave, a bonding
802 * If 'aux' is already registered then this function updates its configuration
803 * to 's'. Otherwise, this function registers a new bundle.
805 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
808 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
809 const struct ofproto_bundle_settings *s)
811 return (ofproto->ofproto_class->bundle_set
812 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
816 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
817 * If no such bundle has been registered, this has no effect. */
819 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
821 return ofproto_bundle_register(ofproto, aux, NULL);
825 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
826 * If 'aux' is already registered then this function updates its configuration
827 * to 's'. Otherwise, this function registers a new mirror. */
829 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
830 const struct ofproto_mirror_settings *s)
832 return (ofproto->ofproto_class->mirror_set
833 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
837 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
838 * If no mirror has been registered, this has no effect. */
840 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
842 return ofproto_mirror_register(ofproto, aux, NULL);
845 /* Retrieves statistics from mirror associated with client data pointer
846 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
847 * 'bytes', respectively. If a particular counters is not supported,
848 * the appropriate argument is set to UINT64_MAX. */
850 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
851 uint64_t *packets, uint64_t *bytes)
853 if (!ofproto->ofproto_class->mirror_get_stats) {
854 *packets = *bytes = UINT64_MAX;
858 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
862 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
863 * which all packets are flooded, instead of using MAC learning. If
864 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
866 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
869 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
871 return (ofproto->ofproto_class->set_flood_vlans
872 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
876 /* Returns true if 'aux' is a registered bundle that is currently in use as the
877 * output for a mirror. */
879 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
881 return (ofproto->ofproto_class->is_mirror_output_bundle
882 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
886 /* Configuration of OpenFlow tables. */
888 /* Returns the number of OpenFlow tables in 'ofproto'. */
890 ofproto_get_n_tables(const struct ofproto *ofproto)
892 return ofproto->n_tables;
895 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
896 * settings from 's'. 'table_id' must be in the range 0 through the number of
897 * OpenFlow tables in 'ofproto' minus 1, inclusive.
899 * For read-only tables, only the name may be configured. */
901 ofproto_configure_table(struct ofproto *ofproto, int table_id,
902 const struct ofproto_table_settings *s)
904 struct oftable *table;
906 assert(table_id >= 0 && table_id < ofproto->n_tables);
907 table = &ofproto->tables[table_id];
909 oftable_set_name(table, s->name);
911 if (table->flags & OFTABLE_READONLY) {
916 oftable_enable_eviction(table, s->groups, s->n_groups);
918 oftable_disable_eviction(table);
921 table->max_flows = s->max_flows;
922 if (classifier_count(&table->cls) > table->max_flows
923 && table->eviction_fields) {
924 /* 'table' contains more flows than allowed. We might not be able to
925 * evict them right away because of the asynchronous nature of flow
926 * table changes. Schedule eviction for later. */
927 switch (ofproto->state) {
929 ofproto->state = S_EVICT;
933 /* We're already deleting flows, nothing more to do. */
940 ofproto_has_snoops(const struct ofproto *ofproto)
942 return connmgr_has_snoops(ofproto->connmgr);
946 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
948 connmgr_get_snoops(ofproto->connmgr, snoops);
952 ofproto_flush__(struct ofproto *ofproto)
954 struct ofopgroup *group;
955 struct oftable *table;
957 if (ofproto->ofproto_class->flush) {
958 ofproto->ofproto_class->flush(ofproto);
961 group = ofopgroup_create_unattached(ofproto);
962 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
963 struct rule *rule, *next_rule;
964 struct cls_cursor cursor;
966 if (table->flags & OFTABLE_HIDDEN) {
970 cls_cursor_init(&cursor, &table->cls, NULL);
971 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
972 if (!rule->pending) {
973 ofoperation_create(group, rule, OFOPERATION_DELETE,
975 oftable_remove_rule(rule);
976 ofproto->ofproto_class->rule_destruct(rule);
980 ofopgroup_submit(group);
984 ofproto_destroy__(struct ofproto *ofproto)
986 struct oftable *table;
988 assert(list_is_empty(&ofproto->pending));
989 assert(!ofproto->n_pending);
991 connmgr_destroy(ofproto->connmgr);
993 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
996 free(ofproto->mfr_desc);
997 free(ofproto->hw_desc);
998 free(ofproto->sw_desc);
999 free(ofproto->serial_desc);
1000 free(ofproto->dp_desc);
1001 hmap_destroy(&ofproto->ports);
1002 shash_destroy(&ofproto->port_by_name);
1003 bitmap_free(ofproto->ofp_port_ids);
1004 simap_destroy(&ofproto->ofp_requests);
1006 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1007 oftable_destroy(table);
1009 free(ofproto->tables);
1011 hmap_destroy(&ofproto->deletions);
1013 free(ofproto->vlan_bitmap);
1015 ofproto->ofproto_class->dealloc(ofproto);
1019 ofproto_destroy(struct ofproto *p)
1021 struct ofport *ofport, *next_ofport;
1028 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1029 ofport_destroy(ofport);
1032 p->ofproto_class->destruct(p);
1033 ofproto_destroy__(p);
1036 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1037 * kernel datapath, for example, this destroys the datapath in the kernel, and
1038 * with the netdev-based datapath, it tears down the data structures that
1039 * represent the datapath.
1041 * The datapath should not be currently open as an ofproto. */
1043 ofproto_delete(const char *name, const char *type)
1045 const struct ofproto_class *class = ofproto_class_find__(type);
1046 return (!class ? EAFNOSUPPORT
1047 : !class->del ? EACCES
1048 : class->del(type, name));
1052 process_port_change(struct ofproto *ofproto, int error, char *devname)
1054 if (error == ENOBUFS) {
1055 reinit_ports(ofproto);
1056 } else if (!error) {
1057 update_port(ofproto, devname);
1063 ofproto_type_run(const char *datapath_type)
1065 const struct ofproto_class *class;
1068 datapath_type = ofproto_normalize_type(datapath_type);
1069 class = ofproto_class_find__(datapath_type);
1071 error = class->type_run ? class->type_run(datapath_type) : 0;
1072 if (error && error != EAGAIN) {
1073 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1074 datapath_type, strerror(error));
1080 ofproto_type_run_fast(const char *datapath_type)
1082 const struct ofproto_class *class;
1085 datapath_type = ofproto_normalize_type(datapath_type);
1086 class = ofproto_class_find__(datapath_type);
1088 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1089 if (error && error != EAGAIN) {
1090 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1091 datapath_type, strerror(error));
1097 ofproto_type_wait(const char *datapath_type)
1099 const struct ofproto_class *class;
1101 datapath_type = ofproto_normalize_type(datapath_type);
1102 class = ofproto_class_find__(datapath_type);
1104 if (class->type_wait) {
1105 class->type_wait(datapath_type);
1110 ofproto_run(struct ofproto *p)
1112 struct sset changed_netdevs;
1113 const char *changed_netdev;
1114 struct ofport *ofport;
1117 error = p->ofproto_class->run(p);
1118 if (error && error != EAGAIN) {
1119 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1122 if (p->ofproto_class->port_poll) {
1125 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1126 process_port_change(p, error, devname);
1130 /* Update OpenFlow port status for any port whose netdev has changed.
1132 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1133 * destroyed, so it's not safe to update ports directly from the
1134 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1135 * need this two-phase approach. */
1136 sset_init(&changed_netdevs);
1137 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1138 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1139 if (ofport->change_seq != change_seq) {
1140 ofport->change_seq = change_seq;
1141 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1144 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1145 update_port(p, changed_netdev);
1147 sset_destroy(&changed_netdevs);
1151 connmgr_run(p->connmgr, handle_openflow);
1155 connmgr_run(p->connmgr, NULL);
1157 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1158 p->state = S_OPENFLOW;
1163 connmgr_run(p->connmgr, NULL);
1165 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1166 connmgr_flushed(p->connmgr);
1167 p->state = S_OPENFLOW;
1175 if (time_msec() >= p->next_op_report) {
1176 long long int ago = (time_msec() - p->first_op) / 1000;
1177 long long int interval = (p->last_op - p->first_op) / 1000;
1181 ds_put_format(&s, "%d flow_mods ",
1182 p->n_add + p->n_delete + p->n_modify);
1183 if (interval == ago) {
1184 ds_put_format(&s, "in the last %lld s", ago);
1185 } else if (interval) {
1186 ds_put_format(&s, "in the %lld s starting %lld s ago",
1189 ds_put_format(&s, "%lld s ago", ago);
1192 ds_put_cstr(&s, " (");
1194 ds_put_format(&s, "%d adds, ", p->n_add);
1197 ds_put_format(&s, "%d deletes, ", p->n_delete);
1200 ds_put_format(&s, "%d modifications, ", p->n_modify);
1203 ds_put_char(&s, ')');
1205 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1208 p->n_add = p->n_delete = p->n_modify = 0;
1209 p->next_op_report = LLONG_MAX;
1215 /* Performs periodic activity required by 'ofproto' that needs to be done
1216 * with the least possible latency.
1218 * It makes sense to call this function a couple of times per poll loop, to
1219 * provide a significant performance boost on some benchmarks with the
1220 * ofproto-dpif implementation. */
1222 ofproto_run_fast(struct ofproto *p)
1226 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1227 if (error && error != EAGAIN) {
1228 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1229 p->name, strerror(error));
1235 ofproto_wait(struct ofproto *p)
1237 struct ofport *ofport;
1239 p->ofproto_class->wait(p);
1240 if (p->ofproto_class->port_poll_wait) {
1241 p->ofproto_class->port_poll_wait(p);
1244 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1245 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1246 poll_immediate_wake();
1252 connmgr_wait(p->connmgr, true);
1257 connmgr_wait(p->connmgr, false);
1258 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1259 poll_immediate_wake();
1266 ofproto_is_alive(const struct ofproto *p)
1268 return connmgr_has_controllers(p->connmgr);
1271 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1272 * memory_report(). */
1274 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1276 const struct oftable *table;
1277 unsigned int n_rules;
1279 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1280 simap_increase(usage, "ops",
1281 ofproto->n_pending + hmap_count(&ofproto->deletions));
1284 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1285 n_rules += classifier_count(&table->cls);
1287 simap_increase(usage, "rules", n_rules);
1289 if (ofproto->ofproto_class->get_memory_usage) {
1290 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1293 connmgr_get_memory_usage(ofproto->connmgr, usage);
1297 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1300 connmgr_get_controller_info(ofproto->connmgr, info);
1304 ofproto_free_ofproto_controller_info(struct shash *info)
1306 connmgr_free_controller_info(info);
1309 /* Makes a deep copy of 'old' into 'port'. */
1311 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1313 port->name = xstrdup(old->name);
1314 port->type = xstrdup(old->type);
1315 port->ofp_port = old->ofp_port;
1318 /* Frees memory allocated to members of 'ofproto_port'.
1320 * Do not call this function on an ofproto_port obtained from
1321 * ofproto_port_dump_next(): that function retains ownership of the data in the
1324 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1326 free(ofproto_port->name);
1327 free(ofproto_port->type);
1330 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1332 * This function provides no status indication. An error status for the entire
1333 * dump operation is provided when it is completed by calling
1334 * ofproto_port_dump_done().
1337 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1338 const struct ofproto *ofproto)
1340 dump->ofproto = ofproto;
1341 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1345 /* Attempts to retrieve another port from 'dump', which must have been created
1346 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1347 * 'port' and returns true. On failure, returns false.
1349 * Failure might indicate an actual error or merely that the last port has been
1350 * dumped. An error status for the entire dump operation is provided when it
1351 * is completed by calling ofproto_port_dump_done().
1353 * The ofproto owns the data stored in 'port'. It will remain valid until at
1354 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1355 * ofproto_port_dump_done(). */
1357 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1358 struct ofproto_port *port)
1360 const struct ofproto *ofproto = dump->ofproto;
1366 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1369 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1375 /* Completes port table dump operation 'dump', which must have been created
1376 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1377 * error-free, otherwise a positive errno value describing the problem. */
1379 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1381 const struct ofproto *ofproto = dump->ofproto;
1383 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1386 return dump->error == EOF ? 0 : dump->error;
1389 /* Returns the type to pass to netdev_open() when a datapath of type
1390 * 'datapath_type' has a port of type 'port_type', for a few special
1391 * cases when a netdev type differs from a port type. For example, when
1392 * using the userspace datapath, a port of type "internal" needs to be
1395 * Returns either 'type' itself or a string literal, which must not be
1398 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1400 const struct ofproto_class *class;
1402 datapath_type = ofproto_normalize_type(datapath_type);
1403 class = ofproto_class_find__(datapath_type);
1408 return (class->port_open_type
1409 ? class->port_open_type(datapath_type, port_type)
1413 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1414 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1415 * the port's OpenFlow port number.
1417 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1418 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1419 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1420 * 'ofp_portp' is non-null). */
1422 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1423 uint16_t *ofp_portp)
1425 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1428 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1430 const char *netdev_name = netdev_get_name(netdev);
1432 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1433 update_port(ofproto, netdev_name);
1436 struct ofproto_port ofproto_port;
1438 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1440 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1441 ofproto_port_destroy(&ofproto_port);
1446 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1447 * initializes '*port' appropriately; on failure, returns a positive errno
1450 * The caller owns the data in 'ofproto_port' and must free it with
1451 * ofproto_port_destroy() when it is no longer needed. */
1453 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1454 struct ofproto_port *port)
1458 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1460 memset(port, 0, sizeof *port);
1465 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1466 * Returns 0 if successful, otherwise a positive errno. */
1468 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1470 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1471 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1472 struct simap_node *ofp_request_node;
1475 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1476 if (ofp_request_node) {
1477 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1480 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1481 if (!error && ofport) {
1482 /* 'name' is the netdev's name and update_port() is going to close the
1483 * netdev. Just in case update_port() refers to 'name' after it
1484 * destroys 'ofport', make a copy of it around the update_port()
1486 char *devname = xstrdup(name);
1487 update_port(ofproto, devname);
1493 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1494 * performs the 'n_actions' actions in 'actions'. The new flow will not
1497 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1498 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1499 * controllers; otherwise, it will be hidden.
1501 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1503 * This is a helper function for in-band control and fail-open. */
1505 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1506 unsigned int priority,
1507 const struct ofpact *ofpacts, size_t ofpacts_len)
1509 const struct rule *rule;
1511 rule = rule_from_cls_rule(classifier_find_match_exactly(
1512 &ofproto->tables[0].cls, match, priority));
1513 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1514 ofpacts, ofpacts_len)) {
1515 struct ofputil_flow_mod fm;
1517 memset(&fm, 0, sizeof fm);
1519 fm.priority = priority;
1520 fm.buffer_id = UINT32_MAX;
1521 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1522 fm.ofpacts_len = ofpacts_len;
1523 add_flow(ofproto, NULL, &fm, NULL);
1528 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1529 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1530 * operation cannot be initiated now but may be retried later.
1532 * This is a helper function for in-band control and fail-open. */
1534 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1536 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1539 /* Searches for a rule with matching criteria exactly equal to 'target' in
1540 * ofproto's table 0 and, if it finds one, deletes it.
1542 * This is a helper function for in-band control and fail-open. */
1544 ofproto_delete_flow(struct ofproto *ofproto,
1545 const struct match *target, unsigned int priority)
1549 rule = rule_from_cls_rule(classifier_find_match_exactly(
1550 &ofproto->tables[0].cls, target, priority));
1552 /* No such rule -> success. */
1554 } else if (rule->pending) {
1555 /* An operation on the rule is already pending -> failure.
1556 * Caller must retry later if it's important. */
1559 /* Initiate deletion -> success. */
1560 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1561 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1562 oftable_remove_rule(rule);
1563 ofproto->ofproto_class->rule_destruct(rule);
1564 ofopgroup_submit(group);
1570 /* Starts the process of deleting all of the flows from all of ofproto's flow
1571 * tables and then reintroducing the flows required by in-band control and
1572 * fail-open. The process will complete in a later call to ofproto_run(). */
1574 ofproto_flush_flows(struct ofproto *ofproto)
1576 COVERAGE_INC(ofproto_flush);
1577 ofproto->state = S_FLUSH;
1581 reinit_ports(struct ofproto *p)
1583 struct ofproto_port_dump dump;
1584 struct sset devnames;
1585 struct ofport *ofport;
1586 struct ofproto_port ofproto_port;
1587 const char *devname;
1589 COVERAGE_INC(ofproto_reinit_ports);
1591 sset_init(&devnames);
1592 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1593 sset_add(&devnames, netdev_get_name(ofport->netdev));
1595 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1596 sset_add(&devnames, ofproto_port.name);
1599 SSET_FOR_EACH (devname, &devnames) {
1600 update_port(p, devname);
1602 sset_destroy(&devnames);
1606 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1610 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1611 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1613 if (ofp_port >= ofproto->max_ports
1614 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1615 bool retry = ofproto->alloc_port_no ? true : false;
1617 /* Search for a free OpenFlow port number. We try not to
1618 * immediately reuse them to prevent problems due to old
1620 while (ofp_port >= ofproto->max_ports) {
1621 for (ofproto->alloc_port_no++;
1622 ofproto->alloc_port_no < ofproto->max_ports;
1623 ofproto->alloc_port_no++) {
1624 if (!bitmap_is_set(ofproto->ofp_port_ids,
1625 ofproto->alloc_port_no)) {
1626 ofp_port = ofproto->alloc_port_no;
1630 if (ofproto->alloc_port_no >= ofproto->max_ports) {
1632 ofproto->alloc_port_no = 0;
1641 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1646 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1648 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1651 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1652 * pointer if the netdev cannot be opened. On success, also fills in
1654 static struct netdev *
1655 ofport_open(struct ofproto *ofproto,
1656 struct ofproto_port *ofproto_port,
1657 struct ofputil_phy_port *pp)
1659 enum netdev_flags flags;
1660 struct netdev *netdev;
1663 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1665 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1666 "cannot be opened (%s)",
1668 ofproto_port->name, ofproto_port->ofp_port,
1669 ofproto_port->name, strerror(error));
1673 if (ofproto_port->ofp_port == OFPP_NONE) {
1674 if (!strcmp(ofproto->name, ofproto_port->name)) {
1675 ofproto_port->ofp_port = OFPP_LOCAL;
1677 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1678 ofproto_port->name);
1681 pp->port_no = ofproto_port->ofp_port;
1682 netdev_get_etheraddr(netdev, pp->hw_addr);
1683 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1684 netdev_get_flags(netdev, &flags);
1685 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1686 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1687 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1688 &pp->supported, &pp->peer);
1689 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1690 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1695 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1696 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1699 ofport_equal(const struct ofputil_phy_port *a,
1700 const struct ofputil_phy_port *b)
1702 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1703 && a->state == b->state
1704 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1705 && a->curr == b->curr
1706 && a->advertised == b->advertised
1707 && a->supported == b->supported
1708 && a->peer == b->peer
1709 && a->curr_speed == b->curr_speed
1710 && a->max_speed == b->max_speed);
1713 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1714 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1715 * one with the same name or port number). */
1717 ofport_install(struct ofproto *p,
1718 struct netdev *netdev, const struct ofputil_phy_port *pp)
1720 const char *netdev_name = netdev_get_name(netdev);
1721 struct ofport *ofport;
1724 /* Create ofport. */
1725 ofport = p->ofproto_class->port_alloc();
1730 ofport->ofproto = p;
1731 ofport->netdev = netdev;
1732 ofport->change_seq = netdev_change_seq(netdev);
1734 ofport->ofp_port = pp->port_no;
1736 /* Add port to 'p'. */
1737 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1738 shash_add(&p->port_by_name, netdev_name, ofport);
1740 update_mtu(p, ofport);
1742 /* Let the ofproto_class initialize its private data. */
1743 error = p->ofproto_class->port_construct(ofport);
1747 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1751 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1752 p->name, netdev_name, strerror(error));
1754 ofport_destroy__(ofport);
1756 netdev_close(netdev);
1760 /* Removes 'ofport' from 'p' and destroys it. */
1762 ofport_remove(struct ofport *ofport)
1764 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1766 ofport_destroy(ofport);
1769 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1772 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1774 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1776 ofport_remove(port);
1780 /* Updates 'port' with new 'pp' description.
1782 * Does not handle a name or port number change. The caller must implement
1783 * such a change as a delete followed by an add. */
1785 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1787 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1788 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1789 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1790 port->pp.state = pp->state;
1791 port->pp.curr = pp->curr;
1792 port->pp.advertised = pp->advertised;
1793 port->pp.supported = pp->supported;
1794 port->pp.peer = pp->peer;
1795 port->pp.curr_speed = pp->curr_speed;
1796 port->pp.max_speed = pp->max_speed;
1798 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1801 /* Update OpenFlow 'state' in 'port' and notify controller. */
1803 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1805 if (port->pp.state != state) {
1806 port->pp.state = state;
1807 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1813 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1815 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1817 if (port->ofproto->ofproto_class->set_realdev) {
1818 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1820 if (port->ofproto->ofproto_class->set_stp_port) {
1821 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1823 if (port->ofproto->ofproto_class->set_cfm) {
1824 port->ofproto->ofproto_class->set_cfm(port, NULL);
1826 if (port->ofproto->ofproto_class->bundle_remove) {
1827 port->ofproto->ofproto_class->bundle_remove(port);
1833 ofport_destroy__(struct ofport *port)
1835 struct ofproto *ofproto = port->ofproto;
1836 const char *name = netdev_get_name(port->netdev);
1838 hmap_remove(&ofproto->ports, &port->hmap_node);
1839 shash_delete(&ofproto->port_by_name,
1840 shash_find(&ofproto->port_by_name, name));
1842 netdev_close(port->netdev);
1843 ofproto->ofproto_class->port_dealloc(port);
1847 ofport_destroy(struct ofport *port)
1850 dealloc_ofp_port(port->ofproto, port->ofp_port);
1851 port->ofproto->ofproto_class->port_destruct(port);
1852 ofport_destroy__(port);
1857 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1859 struct ofport *port;
1861 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1862 hash_int(ofp_port, 0), &ofproto->ports) {
1863 if (port->ofp_port == ofp_port) {
1871 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1873 struct ofproto *ofproto = port->ofproto;
1876 if (ofproto->ofproto_class->port_get_stats) {
1877 error = ofproto->ofproto_class->port_get_stats(port, stats);
1886 update_port(struct ofproto *ofproto, const char *name)
1888 struct ofproto_port ofproto_port;
1889 struct ofputil_phy_port pp;
1890 struct netdev *netdev;
1891 struct ofport *port;
1893 COVERAGE_INC(ofproto_update_port);
1895 /* Fetch 'name''s location and properties from the datapath. */
1896 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1897 ? ofport_open(ofproto, &ofproto_port, &pp)
1900 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1901 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1902 struct netdev *old_netdev = port->netdev;
1904 /* 'name' hasn't changed location. Any properties changed? */
1905 if (!ofport_equal(&port->pp, &pp)) {
1906 ofport_modified(port, &pp);
1909 update_mtu(ofproto, port);
1911 /* Install the newly opened netdev in case it has changed.
1912 * Don't close the old netdev yet in case port_modified has to
1913 * remove a retained reference to it.*/
1914 port->netdev = netdev;
1915 port->change_seq = netdev_change_seq(netdev);
1917 if (port->ofproto->ofproto_class->port_modified) {
1918 port->ofproto->ofproto_class->port_modified(port);
1921 netdev_close(old_netdev);
1923 /* If 'port' is nonnull then its name differs from 'name' and thus
1924 * we should delete it. If we think there's a port named 'name'
1925 * then its port number must be wrong now so delete it too. */
1927 ofport_remove(port);
1929 ofport_remove_with_name(ofproto, name);
1930 ofport_install(ofproto, netdev, &pp);
1933 /* Any port named 'name' is gone now. */
1934 ofport_remove_with_name(ofproto, name);
1936 ofproto_port_destroy(&ofproto_port);
1940 init_ports(struct ofproto *p)
1942 struct ofproto_port_dump dump;
1943 struct ofproto_port ofproto_port;
1944 struct shash_node *node, *next;
1946 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1947 const char *name = ofproto_port.name;
1949 if (shash_find(&p->port_by_name, name)) {
1950 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1953 struct ofputil_phy_port pp;
1954 struct netdev *netdev;
1956 /* Check if an OpenFlow port number had been requested. */
1957 node = shash_find(&init_ofp_ports, name);
1959 const struct iface_hint *iface_hint = node->data;
1960 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
1963 netdev = ofport_open(p, &ofproto_port, &pp);
1965 ofport_install(p, netdev, &pp);
1970 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
1971 struct iface_hint *iface_hint = node->data;
1973 if (!strcmp(iface_hint->br_name, p->name)) {
1974 free(iface_hint->br_name);
1975 free(iface_hint->br_type);
1977 shash_delete(&init_ofp_ports, node);
1984 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1985 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1987 find_min_mtu(struct ofproto *p)
1989 struct ofport *ofport;
1992 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1993 struct netdev *netdev = ofport->netdev;
1996 /* Skip any internal ports, since that's what we're trying to
1998 if (!strcmp(netdev_get_type(netdev), "internal")) {
2002 if (netdev_get_mtu(netdev, &dev_mtu)) {
2005 if (!mtu || dev_mtu < mtu) {
2010 return mtu ? mtu: ETH_PAYLOAD_MAX;
2013 /* Update MTU of all datapath devices on 'p' to the minimum of the
2014 * non-datapath ports in event of 'port' added or changed. */
2016 update_mtu(struct ofproto *p, struct ofport *port)
2018 struct ofport *ofport;
2019 struct netdev *netdev = port->netdev;
2020 int dev_mtu, old_min;
2022 if (netdev_get_mtu(netdev, &dev_mtu)) {
2026 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2027 if (dev_mtu > p->min_mtu) {
2028 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2029 dev_mtu = p->min_mtu;
2032 port->mtu = dev_mtu;
2036 /* For non-internal port find new min mtu. */
2037 old_min = p->min_mtu;
2038 port->mtu = dev_mtu;
2039 p->min_mtu = find_min_mtu(p);
2040 if (p->min_mtu == old_min) {
2044 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2045 struct netdev *netdev = ofport->netdev;
2047 if (!strcmp(netdev_get_type(netdev), "internal")) {
2048 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2049 ofport->mtu = p->min_mtu;
2056 ofproto_rule_destroy__(struct rule *rule)
2059 cls_rule_destroy(&rule->cr);
2060 free(rule->ofpacts);
2061 rule->ofproto->ofproto_class->rule_dealloc(rule);
2065 /* This function allows an ofproto implementation to destroy any rules that
2066 * remain when its ->destruct() function is called. The caller must have
2067 * already uninitialized any derived members of 'rule' (step 5 described in the
2068 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2069 * This function implements steps 6 and 7.
2071 * This function should only be called from an ofproto implementation's
2072 * ->destruct() function. It is not suitable elsewhere. */
2074 ofproto_rule_destroy(struct rule *rule)
2076 assert(!rule->pending);
2077 oftable_remove_rule(rule);
2078 ofproto_rule_destroy__(rule);
2081 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2082 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2084 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2086 return (port == OFPP_ANY
2087 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2090 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2091 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2093 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2095 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2100 case OFOPERATION_ADD:
2101 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2103 case OFOPERATION_DELETE:
2106 case OFOPERATION_MODIFY:
2107 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2113 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2114 * statistics appropriately. 'packet' must have at least sizeof(struct
2115 * ofp10_packet_in) bytes of headroom.
2117 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2118 * with statistics for 'packet' either way.
2120 * Takes ownership of 'packet'. */
2122 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2126 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2128 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2129 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2132 /* Returns true if 'rule' should be hidden from the controller.
2134 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2135 * (e.g. by in-band control) and are intentionally hidden from the
2138 ofproto_rule_is_hidden(const struct rule *rule)
2140 return rule->cr.priority > UINT16_MAX;
2143 static enum oftable_flags
2144 rule_get_flags(const struct rule *rule)
2146 return rule->ofproto->tables[rule->table_id].flags;
2150 rule_is_modifiable(const struct rule *rule)
2152 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2156 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2158 ofconn_send_reply(ofconn, make_echo_reply(oh));
2163 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2165 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2166 struct ofputil_switch_features features;
2167 struct ofport *port;
2173 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2175 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2177 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2178 * if present, are always at the end.) */
2179 n_tables = ofproto->n_tables;
2180 for (i = 0; i < ofproto->n_tables; i++) {
2181 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2187 features.datapath_id = ofproto->datapath_id;
2188 features.n_buffers = pktbuf_capacity();
2189 features.n_tables = n_tables;
2190 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2191 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2193 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2195 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2196 features.auxiliary_id = 0;
2197 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2199 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2200 ofputil_put_switch_features_port(&port->pp, b);
2203 ofconn_send_reply(ofconn, b);
2208 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2210 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2211 struct ofp_switch_config *osc;
2212 enum ofp_config_flags flags;
2216 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2217 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2218 flags = ofproto->frag_handling;
2219 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2220 if (oh->version < OFP13_VERSION
2221 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2222 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2224 osc->flags = htons(flags);
2225 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2226 ofconn_send_reply(ofconn, buf);
2232 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2234 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2235 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2236 uint16_t flags = ntohs(osc->flags);
2238 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2239 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2240 enum ofp_config_flags cur = ofproto->frag_handling;
2241 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2243 assert((cur & OFPC_FRAG_MASK) == cur);
2245 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2246 ofproto->frag_handling = next;
2248 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2250 ofputil_frag_handling_to_string(next));
2254 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2255 ofconn_set_invalid_ttl_to_controller(ofconn,
2256 (oh->version < OFP13_VERSION
2257 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2259 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2264 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2265 * error message code for the caller to propagate upward. Otherwise, returns
2268 * The log message mentions 'msg_type'. */
2270 reject_slave_controller(struct ofconn *ofconn)
2272 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2273 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2274 return OFPERR_OFPBRC_EPERM;
2281 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2283 struct ofproto *p = ofconn_get_ofproto(ofconn);
2284 struct ofputil_packet_out po;
2285 struct ofpbuf *payload;
2286 uint64_t ofpacts_stub[1024 / 8];
2287 struct ofpbuf ofpacts;
2291 COVERAGE_INC(ofproto_packet_out);
2293 error = reject_slave_controller(ofconn);
2298 /* Decode message. */
2299 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2300 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2302 goto exit_free_ofpacts;
2304 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2305 error = OFPERR_OFPBRC_BAD_PORT;
2306 goto exit_free_ofpacts;
2311 if (po.buffer_id != UINT32_MAX) {
2312 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2313 if (error || !payload) {
2314 goto exit_free_ofpacts;
2317 payload = xmalloc(sizeof *payload);
2318 ofpbuf_use_const(payload, po.packet, po.packet_len);
2321 /* Verify actions against packet, then send packet if successful. */
2322 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2323 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2325 error = p->ofproto_class->packet_out(p, payload, &flow,
2326 po.ofpacts, po.ofpacts_len);
2328 ofpbuf_delete(payload);
2331 ofpbuf_uninit(&ofpacts);
2337 update_port_config(struct ofport *port,
2338 enum ofputil_port_config config,
2339 enum ofputil_port_config mask)
2341 enum ofputil_port_config old_config = port->pp.config;
2342 enum ofputil_port_config toggle;
2344 toggle = (config ^ port->pp.config) & mask;
2345 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2346 if (config & OFPUTIL_PC_PORT_DOWN) {
2347 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2349 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2351 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2354 port->pp.config ^= toggle;
2355 if (port->pp.config != old_config) {
2356 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2361 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2363 struct ofproto *p = ofconn_get_ofproto(ofconn);
2364 struct ofputil_port_mod pm;
2365 struct ofport *port;
2368 error = reject_slave_controller(ofconn);
2373 error = ofputil_decode_port_mod(oh, &pm);
2378 port = ofproto_get_port(p, pm.port_no);
2380 return OFPERR_OFPPMFC_BAD_PORT;
2381 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2382 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2384 update_port_config(port, pm.config, pm.mask);
2386 netdev_set_advertisements(port->netdev, pm.advertise);
2393 handle_desc_stats_request(struct ofconn *ofconn,
2394 const struct ofp_header *request)
2396 static const char *default_mfr_desc = "Nicira, Inc.";
2397 static const char *default_hw_desc = "Open vSwitch";
2398 static const char *default_sw_desc = VERSION;
2399 static const char *default_serial_desc = "None";
2400 static const char *default_dp_desc = "None";
2402 struct ofproto *p = ofconn_get_ofproto(ofconn);
2403 struct ofp_desc_stats *ods;
2406 msg = ofpraw_alloc_stats_reply(request, 0);
2407 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2408 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2409 sizeof ods->mfr_desc);
2410 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2411 sizeof ods->hw_desc);
2412 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2413 sizeof ods->sw_desc);
2414 ovs_strlcpy(ods->serial_num,
2415 p->serial_desc ? p->serial_desc : default_serial_desc,
2416 sizeof ods->serial_num);
2417 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2418 sizeof ods->dp_desc);
2419 ofconn_send_reply(ofconn, msg);
2425 handle_table_stats_request(struct ofconn *ofconn,
2426 const struct ofp_header *request)
2428 struct ofproto *p = ofconn_get_ofproto(ofconn);
2429 struct ofp12_table_stats *ots;
2434 /* Set up default values.
2436 * ofp12_table_stats is used as a generic structure as
2437 * it is able to hold all the fields for ofp10_table_stats
2438 * and ofp11_table_stats (and of course itself).
2440 ots = xcalloc(p->n_tables, sizeof *ots);
2441 for (i = 0; i < p->n_tables; i++) {
2442 ots[i].table_id = i;
2443 sprintf(ots[i].name, "table%zu", i);
2444 ots[i].match = htonll(OFPXMT12_MASK);
2445 ots[i].wildcards = htonll(OFPXMT12_MASK);
2446 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2447 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2448 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2449 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2450 ots[i].metadata_match = htonll(UINT64_MAX);
2451 ots[i].metadata_write = htonll(UINT64_MAX);
2452 ots[i].instructions = htonl(OFPIT11_ALL);
2453 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2454 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2455 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2458 p->ofproto_class->get_tables(p, ots);
2460 /* Post-process the tables, dropping hidden tables. */
2461 n_tables = p->n_tables;
2462 for (i = 0; i < p->n_tables; i++) {
2463 const struct oftable *table = &p->tables[i];
2465 if (table->flags & OFTABLE_HIDDEN) {
2471 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2474 if (table->max_flows < ntohl(ots[i].max_entries)) {
2475 ots[i].max_entries = htonl(table->max_flows);
2479 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2480 ofconn_send_reply(ofconn, msg);
2488 append_port_stat(struct ofport *port, struct list *replies)
2490 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2492 /* Intentionally ignore return value, since errors will set
2493 * 'stats' to all-1s, which is correct for OpenFlow, and
2494 * netdev_get_stats() will log errors. */
2495 ofproto_port_get_stats(port, &ops.stats);
2497 ofputil_append_port_stat(replies, &ops);
2501 handle_port_stats_request(struct ofconn *ofconn,
2502 const struct ofp_header *request)
2504 struct ofproto *p = ofconn_get_ofproto(ofconn);
2505 struct ofport *port;
2506 struct list replies;
2510 error = ofputil_decode_port_stats_request(request, &port_no);
2515 ofpmp_init(&replies, request);
2516 if (port_no != OFPP_ANY) {
2517 port = ofproto_get_port(p, port_no);
2519 append_port_stat(port, &replies);
2522 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2523 append_port_stat(port, &replies);
2527 ofconn_send_replies(ofconn, &replies);
2532 handle_port_desc_stats_request(struct ofconn *ofconn,
2533 const struct ofp_header *request)
2535 struct ofproto *p = ofconn_get_ofproto(ofconn);
2536 enum ofp_version version;
2537 struct ofport *port;
2538 struct list replies;
2540 ofpmp_init(&replies, request);
2542 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2543 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2544 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2547 ofconn_send_replies(ofconn, &replies);
2552 calc_flow_duration__(long long int start, long long int now,
2553 uint32_t *sec, uint32_t *nsec)
2555 long long int msecs = now - start;
2556 *sec = msecs / 1000;
2557 *nsec = (msecs % 1000) * (1000 * 1000);
2560 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2561 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2563 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2565 return (table_id == 0xff || table_id < ofproto->n_tables
2567 : OFPERR_OFPBRC_BAD_TABLE_ID);
2571 static struct oftable *
2572 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2574 struct oftable *table;
2576 for (table = &ofproto->tables[table_id];
2577 table < &ofproto->tables[ofproto->n_tables];
2579 if (!(table->flags & OFTABLE_HIDDEN)) {
2587 static struct oftable *
2588 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2590 if (table_id == 0xff) {
2591 return next_visible_table(ofproto, 0);
2592 } else if (table_id < ofproto->n_tables) {
2593 return &ofproto->tables[table_id];
2599 static struct oftable *
2600 next_matching_table(const struct ofproto *ofproto,
2601 const struct oftable *table, uint8_t table_id)
2603 return (table_id == 0xff
2604 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2608 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2610 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2611 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2613 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2614 * only once, for that table. (This can be used to access tables marked
2617 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2618 * entered at all. (Perhaps you should have validated TABLE_ID with
2619 * check_table_id().)
2621 * All parameters are evaluated multiple times.
2623 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2624 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2626 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2628 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2629 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2630 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2633 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2634 * to 'out_port' are included.
2636 * Hidden rules are always omitted.
2638 * Returns 0 on success, otherwise an OpenFlow error code. */
2640 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2641 const struct match *match,
2642 ovs_be64 cookie, ovs_be64 cookie_mask,
2643 uint16_t out_port, struct list *rules)
2645 struct oftable *table;
2649 error = check_table_id(ofproto, table_id);
2655 cls_rule_init(&cr, match, 0);
2656 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2657 struct cls_cursor cursor;
2660 cls_cursor_init(&cursor, &table->cls, &cr);
2661 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2662 if (rule->pending) {
2663 error = OFPROTO_POSTPONE;
2666 if (!ofproto_rule_is_hidden(rule)
2667 && ofproto_rule_has_out_port(rule, out_port)
2668 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2669 list_push_back(rules, &rule->ofproto_node);
2675 cls_rule_destroy(&cr);
2679 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2680 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2681 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2684 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2685 * to 'out_port' are included.
2687 * Hidden rules are always omitted.
2689 * Returns 0 on success, otherwise an OpenFlow error code. */
2691 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2692 const struct match *match, unsigned int priority,
2693 ovs_be64 cookie, ovs_be64 cookie_mask,
2694 uint16_t out_port, struct list *rules)
2696 struct oftable *table;
2700 error = check_table_id(ofproto, table_id);
2706 cls_rule_init(&cr, match, priority);
2707 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2710 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2713 if (rule->pending) {
2714 error = OFPROTO_POSTPONE;
2717 if (!ofproto_rule_is_hidden(rule)
2718 && ofproto_rule_has_out_port(rule, out_port)
2719 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2720 list_push_back(rules, &rule->ofproto_node);
2726 cls_rule_destroy(&cr);
2730 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2731 * forced into the range of a uint16_t. */
2733 age_secs(long long int age_ms)
2735 return (age_ms < 0 ? 0
2736 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2737 : (unsigned int) age_ms / 1000);
2741 handle_flow_stats_request(struct ofconn *ofconn,
2742 const struct ofp_header *request)
2744 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2745 struct ofputil_flow_stats_request fsr;
2746 struct list replies;
2751 error = ofputil_decode_flow_stats_request(&fsr, request);
2756 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2757 fsr.cookie, fsr.cookie_mask,
2758 fsr.out_port, &rules);
2763 ofpmp_init(&replies, request);
2764 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2765 long long int now = time_msec();
2766 struct ofputil_flow_stats fs;
2768 minimatch_expand(&rule->cr.match, &fs.match);
2769 fs.priority = rule->cr.priority;
2770 fs.cookie = rule->flow_cookie;
2771 fs.table_id = rule->table_id;
2772 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2774 fs.idle_timeout = rule->idle_timeout;
2775 fs.hard_timeout = rule->hard_timeout;
2776 fs.idle_age = age_secs(now - rule->used);
2777 fs.hard_age = age_secs(now - rule->modified);
2778 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2780 fs.ofpacts = rule->ofpacts;
2781 fs.ofpacts_len = rule->ofpacts_len;
2783 if (rule->send_flow_removed) {
2784 fs.flags |= OFPFF_SEND_FLOW_REM;
2785 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2786 and OFPFF13_NO_BYT_COUNTS */
2788 ofputil_append_flow_stats_reply(&fs, &replies);
2790 ofconn_send_replies(ofconn, &replies);
2796 flow_stats_ds(struct rule *rule, struct ds *results)
2798 uint64_t packet_count, byte_count;
2800 rule->ofproto->ofproto_class->rule_get_stats(rule,
2801 &packet_count, &byte_count);
2803 if (rule->table_id != 0) {
2804 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2806 ds_put_format(results, "duration=%llds, ",
2807 (time_msec() - rule->created) / 1000);
2808 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2809 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2810 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2811 cls_rule_format(&rule->cr, results);
2812 ds_put_char(results, ',');
2813 if (rule->ofpacts_len > 0) {
2814 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2816 ds_put_cstr(results, "drop");
2818 ds_put_cstr(results, "\n");
2821 /* Adds a pretty-printed description of all flows to 'results', including
2822 * hidden flows (e.g., set up by in-band control). */
2824 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2826 struct oftable *table;
2828 OFPROTO_FOR_EACH_TABLE (table, p) {
2829 struct cls_cursor cursor;
2832 cls_cursor_init(&cursor, &table->cls, NULL);
2833 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2834 flow_stats_ds(rule, results);
2839 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2840 * '*engine_type' and '*engine_id', respectively. */
2842 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2843 uint8_t *engine_type, uint8_t *engine_id)
2845 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2848 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2849 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2850 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2851 * and -1 if CFM is not enabled on 'ofp_port'. */
2853 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2855 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2856 return (ofport && ofproto->ofproto_class->get_cfm_fault
2857 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2861 /* Checks the operational status reported by the remote CFM endpoint of
2862 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2863 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2866 ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
2868 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2869 return (ofport && ofproto->ofproto_class->get_cfm_opup
2870 ? ofproto->ofproto_class->get_cfm_opup(ofport)
2874 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2875 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2876 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2877 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2879 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2880 uint16_t ofp_port, const uint64_t **rmps,
2883 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2887 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2888 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2893 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2894 * integer value between 0 and 100 to indicate the health of the port as a
2895 * percentage which is the average of cfm health of all the remote_mpids or
2896 * returns -1 if CFM is not enabled on 'ofport'. */
2898 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2900 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2901 return (ofport && ofproto->ofproto_class->get_cfm_health
2902 ? ofproto->ofproto_class->get_cfm_health(ofport)
2907 handle_aggregate_stats_request(struct ofconn *ofconn,
2908 const struct ofp_header *oh)
2910 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2911 struct ofputil_flow_stats_request request;
2912 struct ofputil_aggregate_stats stats;
2913 bool unknown_packets, unknown_bytes;
2914 struct ofpbuf *reply;
2919 error = ofputil_decode_flow_stats_request(&request, oh);
2924 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2925 request.cookie, request.cookie_mask,
2926 request.out_port, &rules);
2931 memset(&stats, 0, sizeof stats);
2932 unknown_packets = unknown_bytes = false;
2933 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2934 uint64_t packet_count;
2935 uint64_t byte_count;
2937 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2940 if (packet_count == UINT64_MAX) {
2941 unknown_packets = true;
2943 stats.packet_count += packet_count;
2946 if (byte_count == UINT64_MAX) {
2947 unknown_bytes = true;
2949 stats.byte_count += byte_count;
2954 if (unknown_packets) {
2955 stats.packet_count = UINT64_MAX;
2957 if (unknown_bytes) {
2958 stats.byte_count = UINT64_MAX;
2961 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2962 ofconn_send_reply(ofconn, reply);
2967 struct queue_stats_cbdata {
2968 struct ofport *ofport;
2969 struct list replies;
2973 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2974 const struct netdev_queue_stats *stats)
2977 struct ofputil_queue_stats oqs = {
2978 .port_no = cbdata->ofport->pp.port_no,
2979 .queue_id = queue_id,
2982 ofputil_append_queue_stat(&cbdata->replies, &oqs);
2986 handle_queue_stats_dump_cb(uint32_t queue_id,
2987 struct netdev_queue_stats *stats,
2990 struct queue_stats_cbdata *cbdata = cbdata_;
2992 put_queue_stats(cbdata, queue_id, stats);
2996 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2997 struct queue_stats_cbdata *cbdata)
2999 cbdata->ofport = port;
3000 if (queue_id == OFPQ_ALL) {
3001 netdev_dump_queue_stats(port->netdev,
3002 handle_queue_stats_dump_cb, cbdata);
3004 struct netdev_queue_stats stats;
3006 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3007 put_queue_stats(cbdata, queue_id, &stats);
3009 return OFPERR_OFPQOFC_BAD_QUEUE;
3016 handle_queue_stats_request(struct ofconn *ofconn,
3017 const struct ofp_header *rq)
3019 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3020 struct queue_stats_cbdata cbdata;
3021 struct ofport *port;
3023 struct ofputil_queue_stats_request oqsr;
3025 COVERAGE_INC(ofproto_queue_req);
3027 ofpmp_init(&cbdata.replies, rq);
3029 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3034 if (oqsr.port_no == OFPP_ANY) {
3035 error = OFPERR_OFPQOFC_BAD_QUEUE;
3036 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3037 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3042 port = ofproto_get_port(ofproto, oqsr.port_no);
3044 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3045 : OFPERR_OFPQOFC_BAD_PORT);
3048 ofconn_send_replies(ofconn, &cbdata.replies);
3050 ofpbuf_list_delete(&cbdata.replies);
3057 is_flow_deletion_pending(const struct ofproto *ofproto,
3058 const struct cls_rule *cls_rule,
3061 if (!hmap_is_empty(&ofproto->deletions)) {
3062 struct ofoperation *op;
3064 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3065 cls_rule_hash(cls_rule, table_id),
3066 &ofproto->deletions) {
3067 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3076 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3077 * in which no matching flow already exists in the flow table.
3079 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3080 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3081 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3082 * initiated now but may be retried later.
3084 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3085 * ownership remains with the caller.
3087 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3090 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3091 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3093 struct oftable *table;
3094 struct ofopgroup *group;
3095 struct rule *victim;
3100 error = check_table_id(ofproto, fm->table_id);
3106 if (fm->table_id == 0xff) {
3108 if (ofproto->ofproto_class->rule_choose_table) {
3109 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3115 assert(table_id < ofproto->n_tables);
3116 table = &ofproto->tables[table_id];
3118 table = &ofproto->tables[0];
3120 } else if (fm->table_id < ofproto->n_tables) {
3121 table = &ofproto->tables[fm->table_id];
3123 return OFPERR_OFPBRC_BAD_TABLE_ID;
3126 if (table->flags & OFTABLE_READONLY) {
3127 return OFPERR_OFPBRC_EPERM;
3130 /* Allocate new rule and initialize classifier rule. */
3131 rule = ofproto->ofproto_class->rule_alloc();
3133 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3134 ofproto->name, strerror(error));
3137 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3139 /* Serialize against pending deletion. */
3140 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3141 cls_rule_destroy(&rule->cr);
3142 ofproto->ofproto_class->rule_dealloc(rule);
3143 return OFPROTO_POSTPONE;
3146 /* Check for overlap, if requested. */
3147 if (fm->flags & OFPFF_CHECK_OVERLAP
3148 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3149 cls_rule_destroy(&rule->cr);
3150 ofproto->ofproto_class->rule_dealloc(rule);
3151 return OFPERR_OFPFMFC_OVERLAP;
3154 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3156 rule->ofproto = ofproto;
3157 rule->pending = NULL;
3158 rule->flow_cookie = fm->new_cookie;
3159 rule->created = rule->modified = rule->used = time_msec();
3160 rule->idle_timeout = fm->idle_timeout;
3161 rule->hard_timeout = fm->hard_timeout;
3162 rule->table_id = table - ofproto->tables;
3163 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3164 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3165 and OFPFF13_NO_BYT_COUNTS */
3166 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3167 rule->ofpacts_len = fm->ofpacts_len;
3168 rule->evictable = true;
3169 rule->eviction_group = NULL;
3170 rule->monitor_flags = 0;
3171 rule->add_seqno = 0;
3172 rule->modify_seqno = 0;
3174 /* Insert new rule. */
3175 victim = oftable_replace_rule(rule);
3176 if (victim && !rule_is_modifiable(victim)) {
3177 error = OFPERR_OFPBRC_EPERM;
3178 } else if (victim && victim->pending) {
3179 error = OFPROTO_POSTPONE;
3181 struct ofoperation *op;
3184 if (classifier_count(&table->cls) > table->max_flows) {
3187 was_evictable = rule->evictable;
3188 rule->evictable = false;
3189 evict = choose_rule_to_evict(table);
3190 rule->evictable = was_evictable;
3193 error = OFPERR_OFPFMFC_TABLE_FULL;
3195 } else if (evict->pending) {
3196 error = OFPROTO_POSTPONE;
3203 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3204 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3205 op->victim = victim;
3207 error = ofproto->ofproto_class->rule_construct(rule);
3209 op->group->n_running--;
3210 ofoperation_destroy(rule->pending);
3212 delete_flow__(evict, group);
3214 ofopgroup_submit(group);
3218 /* Back out if an error occurred. */
3220 oftable_substitute_rule(rule, victim);
3221 ofproto_rule_destroy__(rule);
3226 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3228 /* Modifies the rules listed in 'rules', changing their actions to match those
3231 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3234 * Returns 0 on success, otherwise an OpenFlow error code. */
3236 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3237 const struct ofputil_flow_mod *fm,
3238 const struct ofp_header *request, struct list *rules)
3240 struct ofopgroup *group;
3244 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3245 error = OFPERR_OFPBRC_EPERM;
3246 LIST_FOR_EACH (rule, ofproto_node, rules) {
3247 struct ofoperation *op;
3248 bool actions_changed;
3249 ovs_be64 new_cookie;
3251 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3253 if (rule_is_modifiable(rule)) {
3254 /* At least one rule is modifiable, don't report EPERM error. */
3260 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3261 rule->ofpacts, rule->ofpacts_len);
3262 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3264 : rule->flow_cookie);
3265 if (!actions_changed && new_cookie == rule->flow_cookie) {
3266 /* No change at all. */
3270 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3271 rule->flow_cookie = new_cookie;
3272 if (actions_changed) {
3273 op->ofpacts = rule->ofpacts;
3274 op->ofpacts_len = rule->ofpacts_len;
3275 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3276 rule->ofpacts_len = fm->ofpacts_len;
3277 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3279 ofoperation_complete(op, 0);
3282 ofopgroup_submit(group);
3288 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3289 const struct ofputil_flow_mod *fm,
3290 const struct ofp_header *request)
3292 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3295 return add_flow(ofproto, ofconn, fm, request);
3298 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3301 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3304 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3305 const struct ofputil_flow_mod *fm,
3306 const struct ofp_header *request)
3311 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3312 fm->cookie, fm->cookie_mask,
3316 } else if (list_is_empty(&rules)) {
3317 return modify_flows_add(ofproto, ofconn, fm, request);
3319 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3323 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3326 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3329 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3330 const struct ofputil_flow_mod *fm,
3331 const struct ofp_header *request)
3336 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3337 fm->priority, fm->cookie, fm->cookie_mask,
3342 } else if (list_is_empty(&rules)) {
3343 return modify_flows_add(ofproto, ofconn, fm, request);
3345 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3346 fm, request, &rules)
3351 /* OFPFC_DELETE implementation. */
3354 delete_flow__(struct rule *rule, struct ofopgroup *group)
3356 struct ofproto *ofproto = rule->ofproto;
3358 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3360 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3361 oftable_remove_rule(rule);
3362 ofproto->ofproto_class->rule_destruct(rule);
3365 /* Deletes the rules listed in 'rules'.
3367 * Returns 0 on success, otherwise an OpenFlow error code. */
3369 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3370 const struct ofp_header *request, struct list *rules)
3372 struct rule *rule, *next;
3373 struct ofopgroup *group;
3375 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3376 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3377 delete_flow__(rule, group);
3379 ofopgroup_submit(group);
3384 /* Implements OFPFC_DELETE. */
3386 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3387 const struct ofputil_flow_mod *fm,
3388 const struct ofp_header *request)
3393 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3394 fm->cookie, fm->cookie_mask,
3395 fm->out_port, &rules);
3396 return (error ? error
3397 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3402 /* Implements OFPFC_DELETE_STRICT. */
3404 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3405 const struct ofputil_flow_mod *fm,
3406 const struct ofp_header *request)
3411 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3412 fm->priority, fm->cookie, fm->cookie_mask,
3413 fm->out_port, &rules);
3414 return (error ? error
3415 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3421 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3423 struct ofputil_flow_removed fr;
3425 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3429 minimatch_expand(&rule->cr.match, &fr.match);
3430 fr.priority = rule->cr.priority;
3431 fr.cookie = rule->flow_cookie;
3433 fr.table_id = rule->table_id;
3434 calc_flow_duration__(rule->created, time_msec(),
3435 &fr.duration_sec, &fr.duration_nsec);
3436 fr.idle_timeout = rule->idle_timeout;
3437 fr.hard_timeout = rule->hard_timeout;
3438 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3441 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3445 ofproto_rule_update_used(struct rule *rule, long long int used)
3447 if (used > rule->used) {
3448 struct eviction_group *evg = rule->eviction_group;
3452 heap_change(&evg->rules, &rule->evg_node,
3453 rule_eviction_priority(rule));
3458 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3459 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3462 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3465 * ofproto implementation ->run() functions should use this function to expire
3466 * OpenFlow flows. */
3468 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3470 struct ofproto *ofproto = rule->ofproto;
3471 struct ofopgroup *group;
3473 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3475 ofproto_rule_send_removed(rule, reason);
3477 group = ofopgroup_create_unattached(ofproto);
3478 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3479 oftable_remove_rule(rule);
3480 ofproto->ofproto_class->rule_destruct(rule);
3481 ofopgroup_submit(group);
3485 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3487 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3488 struct ofputil_flow_mod fm;
3489 uint64_t ofpacts_stub[1024 / 8];
3490 struct ofpbuf ofpacts;
3494 error = reject_slave_controller(ofconn);
3499 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3500 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3503 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3504 &fm.match.flow, ofproto->max_ports);
3507 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3510 goto exit_free_ofpacts;
3513 /* Record the operation for logging a summary report. */
3514 switch (fm.command) {
3520 case OFPFC_MODIFY_STRICT:
3521 ofproto->n_modify++;
3525 case OFPFC_DELETE_STRICT:
3526 ofproto->n_delete++;
3531 if (ofproto->next_op_report == LLONG_MAX) {
3532 ofproto->first_op = now;
3533 ofproto->next_op_report = MAX(now + 10 * 1000,
3534 ofproto->op_backoff);
3535 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3537 ofproto->last_op = now;
3540 ofpbuf_uninit(&ofpacts);
3546 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3547 const struct ofputil_flow_mod *fm,
3548 const struct ofp_header *oh)
3550 if (ofproto->n_pending >= 50) {
3551 assert(!list_is_empty(&ofproto->pending));
3552 return OFPROTO_POSTPONE;
3555 switch (fm->command) {
3557 return add_flow(ofproto, ofconn, fm, oh);
3560 return modify_flows_loose(ofproto, ofconn, fm, oh);
3562 case OFPFC_MODIFY_STRICT:
3563 return modify_flow_strict(ofproto, ofconn, fm, oh);
3566 return delete_flows_loose(ofproto, ofconn, fm, oh);
3568 case OFPFC_DELETE_STRICT:
3569 return delete_flow_strict(ofproto, ofconn, fm, oh);
3572 if (fm->command > 0xff) {
3573 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3574 "flow_mod_table_id extension is not enabled",
3577 return OFPERR_OFPFMFC_BAD_COMMAND;
3582 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3584 const struct nx_role_request *nrr = ofpmsg_body(oh);
3585 struct nx_role_request *reply;
3589 role = ntohl(nrr->role);
3590 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3591 && role != NX_ROLE_SLAVE) {
3592 return OFPERR_OFPRRFC_BAD_ROLE;
3595 if (ofconn_get_role(ofconn) != role
3596 && ofconn_has_pending_opgroups(ofconn)) {
3597 return OFPROTO_POSTPONE;
3600 ofconn_set_role(ofconn, role);
3602 buf = ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY, oh, 0);
3603 reply = ofpbuf_put_zeros(buf, sizeof *reply);
3604 reply->role = htonl(role);
3605 ofconn_send_reply(ofconn, buf);
3611 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3612 const struct ofp_header *oh)
3614 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3615 enum ofputil_protocol cur, next;
3617 cur = ofconn_get_protocol(ofconn);
3618 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3619 ofconn_set_protocol(ofconn, next);
3625 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3627 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3628 enum ofputil_protocol cur, next;
3629 enum ofputil_protocol next_base;
3631 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3633 return OFPERR_OFPBRC_EPERM;
3636 cur = ofconn_get_protocol(ofconn);
3637 next = ofputil_protocol_set_base(cur, next_base);
3638 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3639 /* Avoid sending async messages in surprising protocol. */
3640 return OFPROTO_POSTPONE;
3643 ofconn_set_protocol(ofconn, next);
3648 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3649 const struct ofp_header *oh)
3651 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3654 format = ntohl(msg->format);
3655 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3656 return OFPERR_OFPBRC_EPERM;
3659 if (format != ofconn_get_packet_in_format(ofconn)
3660 && ofconn_has_pending_opgroups(ofconn)) {
3661 /* Avoid sending async message in surprsing packet in format. */
3662 return OFPROTO_POSTPONE;
3665 ofconn_set_packet_in_format(ofconn, format);
3670 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3672 const struct nx_async_config *msg = ofpmsg_body(oh);
3673 uint32_t master[OAM_N_TYPES];
3674 uint32_t slave[OAM_N_TYPES];
3676 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3677 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3678 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3680 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3681 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3682 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3684 ofconn_set_async_config(ofconn, master, slave);
3685 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3686 !ofconn_get_miss_send_len(ofconn)) {
3687 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3694 handle_nxt_set_controller_id(struct ofconn *ofconn,
3695 const struct ofp_header *oh)
3697 const struct nx_controller_id *nci = ofpmsg_body(oh);
3699 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3700 return OFPERR_NXBRC_MUST_BE_ZERO;
3703 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3708 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3712 if (ofconn_has_pending_opgroups(ofconn)) {
3713 return OFPROTO_POSTPONE;
3716 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3717 ? OFPRAW_OFPT10_BARRIER_REPLY
3718 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3719 ofconn_send_reply(ofconn, buf);
3724 ofproto_compose_flow_refresh_update(const struct rule *rule,
3725 enum nx_flow_monitor_flags flags,
3728 struct ofoperation *op = rule->pending;
3729 struct ofputil_flow_update fu;
3732 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3733 /* We'll report the final flow when the operation completes. Reporting
3734 * it now would cause a duplicate report later. */
3738 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3739 ? NXFME_ADDED : NXFME_MODIFIED);
3741 fu.idle_timeout = rule->idle_timeout;
3742 fu.hard_timeout = rule->hard_timeout;
3743 fu.table_id = rule->table_id;
3744 fu.cookie = rule->flow_cookie;
3745 minimatch_expand(&rule->cr.match, &match);
3747 fu.priority = rule->cr.priority;
3748 if (!(flags & NXFMF_ACTIONS)) {
3752 fu.ofpacts = rule->ofpacts;
3753 fu.ofpacts_len = rule->ofpacts_len;
3755 /* An operation is in progress. Use the previous version of the flow's
3756 * actions, so that when the operation commits we report the change. */
3758 case OFOPERATION_ADD:
3759 /* We already verified that there was a victim. */
3760 fu.ofpacts = op->victim->ofpacts;
3761 fu.ofpacts_len = op->victim->ofpacts_len;
3764 case OFOPERATION_MODIFY:
3766 fu.ofpacts = op->ofpacts;
3767 fu.ofpacts_len = op->ofpacts_len;
3769 fu.ofpacts = rule->ofpacts;
3770 fu.ofpacts_len = rule->ofpacts_len;
3774 case OFOPERATION_DELETE:
3775 fu.ofpacts = rule->ofpacts;
3776 fu.ofpacts_len = rule->ofpacts_len;
3784 if (list_is_empty(msgs)) {
3785 ofputil_start_flow_update(msgs);
3787 ofputil_append_flow_update(&fu, msgs);
3791 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3795 LIST_FOR_EACH (rule, ofproto_node, rules) {
3796 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3797 rule->monitor_flags = 0;
3799 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3804 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3805 struct rule *rule, uint64_t seqno,
3808 enum nx_flow_monitor_flags update;
3810 if (ofproto_rule_is_hidden(rule)) {
3815 ? ofoperation_has_out_port(rule->pending, m->out_port)
3816 : ofproto_rule_has_out_port(rule, m->out_port))) {
3821 if (rule->add_seqno > seqno) {
3822 update = NXFMF_ADD | NXFMF_MODIFY;
3823 } else if (rule->modify_seqno > seqno) {
3824 update = NXFMF_MODIFY;
3829 if (!(m->flags & update)) {
3833 update = NXFMF_INITIAL;
3836 if (!rule->monitor_flags) {
3837 list_push_back(rules, &rule->ofproto_node);
3839 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3843 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3847 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3848 const struct ofoperation *op;
3849 const struct oftable *table;
3850 struct cls_rule target;
3852 cls_rule_init_from_minimatch(&target, &m->match, 0);
3853 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3854 struct cls_cursor cursor;
3857 cls_cursor_init(&cursor, &table->cls, &target);
3858 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3859 assert(!rule->pending); /* XXX */
3860 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3864 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3865 struct rule *rule = op->rule;
3867 if (((m->table_id == 0xff
3868 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3869 : m->table_id == rule->table_id))
3870 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3871 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3874 cls_rule_destroy(&target);
3878 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3881 if (m->flags & NXFMF_INITIAL) {
3882 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3887 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3888 uint64_t seqno, struct list *rules)
3890 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3894 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3896 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3897 struct ofmonitor **monitors;
3898 size_t n_monitors, allocated_monitors;
3899 struct list replies;
3906 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3908 n_monitors = allocated_monitors = 0;
3910 struct ofputil_flow_monitor_request request;
3911 struct ofmonitor *m;
3914 retval = ofputil_decode_flow_monitor_request(&request, &b);
3915 if (retval == EOF) {
3917 } else if (retval) {
3922 if (request.table_id != 0xff
3923 && request.table_id >= ofproto->n_tables) {
3924 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3928 error = ofmonitor_create(&request, ofconn, &m);
3933 if (n_monitors >= allocated_monitors) {
3934 monitors = x2nrealloc(monitors, &allocated_monitors,
3937 monitors[n_monitors++] = m;
3941 for (i = 0; i < n_monitors; i++) {
3942 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3945 ofpmp_init(&replies, oh);
3946 ofmonitor_compose_refresh_updates(&rules, &replies);
3947 ofconn_send_replies(ofconn, &replies);
3954 for (i = 0; i < n_monitors; i++) {
3955 ofmonitor_destroy(monitors[i]);
3962 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3964 struct ofmonitor *m;
3967 id = ofputil_decode_flow_monitor_cancel(oh);
3968 m = ofmonitor_lookup(ofconn, id);
3970 return OFPERR_NXBRC_FM_BAD_ID;
3973 ofmonitor_destroy(m);
3978 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3980 const struct ofp_header *oh = msg->data;
3984 error = ofptype_decode(&type, oh);
3990 /* OpenFlow requests. */
3991 case OFPTYPE_ECHO_REQUEST:
3992 return handle_echo_request(ofconn, oh);
3994 case OFPTYPE_FEATURES_REQUEST:
3995 return handle_features_request(ofconn, oh);
3997 case OFPTYPE_GET_CONFIG_REQUEST:
3998 return handle_get_config_request(ofconn, oh);
4000 case OFPTYPE_SET_CONFIG:
4001 return handle_set_config(ofconn, oh);
4003 case OFPTYPE_PACKET_OUT:
4004 return handle_packet_out(ofconn, oh);
4006 case OFPTYPE_PORT_MOD:
4007 return handle_port_mod(ofconn, oh);
4009 case OFPTYPE_FLOW_MOD:
4010 return handle_flow_mod(ofconn, oh);
4012 case OFPTYPE_BARRIER_REQUEST:
4013 return handle_barrier_request(ofconn, oh);
4015 /* OpenFlow replies. */
4016 case OFPTYPE_ECHO_REPLY:
4019 /* Nicira extension requests. */
4020 case OFPTYPE_ROLE_REQUEST:
4021 return handle_role_request(ofconn, oh);
4023 case OFPTYPE_FLOW_MOD_TABLE_ID:
4024 return handle_nxt_flow_mod_table_id(ofconn, oh);
4026 case OFPTYPE_SET_FLOW_FORMAT:
4027 return handle_nxt_set_flow_format(ofconn, oh);
4029 case OFPTYPE_SET_PACKET_IN_FORMAT:
4030 return handle_nxt_set_packet_in_format(ofconn, oh);
4032 case OFPTYPE_SET_CONTROLLER_ID:
4033 return handle_nxt_set_controller_id(ofconn, oh);
4035 case OFPTYPE_FLOW_AGE:
4036 /* Nothing to do. */
4039 case OFPTYPE_FLOW_MONITOR_CANCEL:
4040 return handle_flow_monitor_cancel(ofconn, oh);
4042 case OFPTYPE_SET_ASYNC_CONFIG:
4043 return handle_nxt_set_async_config(ofconn, oh);
4045 /* Statistics requests. */
4046 case OFPTYPE_DESC_STATS_REQUEST:
4047 return handle_desc_stats_request(ofconn, oh);
4049 case OFPTYPE_FLOW_STATS_REQUEST:
4050 return handle_flow_stats_request(ofconn, oh);
4052 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4053 return handle_aggregate_stats_request(ofconn, oh);
4055 case OFPTYPE_TABLE_STATS_REQUEST:
4056 return handle_table_stats_request(ofconn, oh);
4058 case OFPTYPE_PORT_STATS_REQUEST:
4059 return handle_port_stats_request(ofconn, oh);
4061 case OFPTYPE_QUEUE_STATS_REQUEST:
4062 return handle_queue_stats_request(ofconn, oh);
4064 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4065 return handle_port_desc_stats_request(ofconn, oh);
4067 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4068 return handle_flow_monitor_request(ofconn, oh);
4070 /* FIXME: Change the following once they are implemented: */
4071 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4072 case OFPTYPE_GET_ASYNC_REQUEST:
4073 case OFPTYPE_METER_MOD:
4074 case OFPTYPE_GROUP_REQUEST:
4075 case OFPTYPE_GROUP_DESC_REQUEST:
4076 case OFPTYPE_GROUP_FEATURES_REQUEST:
4077 case OFPTYPE_METER_REQUEST:
4078 case OFPTYPE_METER_CONFIG_REQUEST:
4079 case OFPTYPE_METER_FEATURES_REQUEST:
4080 case OFPTYPE_TABLE_FEATURES_REQUEST:
4081 return OFPERR_OFPBRC_BAD_TYPE;
4085 case OFPTYPE_FEATURES_REPLY:
4086 case OFPTYPE_GET_CONFIG_REPLY:
4087 case OFPTYPE_PACKET_IN:
4088 case OFPTYPE_FLOW_REMOVED:
4089 case OFPTYPE_PORT_STATUS:
4090 case OFPTYPE_BARRIER_REPLY:
4091 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4092 case OFPTYPE_DESC_STATS_REPLY:
4093 case OFPTYPE_FLOW_STATS_REPLY:
4094 case OFPTYPE_QUEUE_STATS_REPLY:
4095 case OFPTYPE_PORT_STATS_REPLY:
4096 case OFPTYPE_TABLE_STATS_REPLY:
4097 case OFPTYPE_AGGREGATE_STATS_REPLY:
4098 case OFPTYPE_PORT_DESC_STATS_REPLY:
4099 case OFPTYPE_ROLE_REPLY:
4100 case OFPTYPE_FLOW_MONITOR_PAUSED:
4101 case OFPTYPE_FLOW_MONITOR_RESUMED:
4102 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4103 case OFPTYPE_GET_ASYNC_REPLY:
4104 case OFPTYPE_GROUP_REPLY:
4105 case OFPTYPE_GROUP_DESC_REPLY:
4106 case OFPTYPE_GROUP_FEATURES_REPLY:
4107 case OFPTYPE_METER_REPLY:
4108 case OFPTYPE_METER_CONFIG_REPLY:
4109 case OFPTYPE_METER_FEATURES_REPLY:
4110 case OFPTYPE_TABLE_FEATURES_REPLY:
4112 return OFPERR_OFPBRC_BAD_TYPE;
4117 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4119 int error = handle_openflow__(ofconn, ofp_msg);
4120 if (error && error != OFPROTO_POSTPONE) {
4121 ofconn_send_error(ofconn, ofp_msg->data, error);
4123 COVERAGE_INC(ofproto_recv_openflow);
4124 return error != OFPROTO_POSTPONE;
4127 /* Asynchronous operations. */
4129 /* Creates and returns a new ofopgroup that is not associated with any
4130 * OpenFlow connection.
4132 * The caller should add operations to the returned group with
4133 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4134 static struct ofopgroup *
4135 ofopgroup_create_unattached(struct ofproto *ofproto)
4137 struct ofopgroup *group = xzalloc(sizeof *group);
4138 group->ofproto = ofproto;
4139 list_init(&group->ofproto_node);
4140 list_init(&group->ops);
4141 list_init(&group->ofconn_node);
4145 /* Creates and returns a new ofopgroup for 'ofproto'.
4147 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4148 * connection. The 'request' and 'buffer_id' arguments are ignored.
4150 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4151 * If the ofopgroup eventually fails, then the error reply will include
4152 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4153 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4155 * The caller should add operations to the returned group with
4156 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4157 static struct ofopgroup *
4158 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4159 const struct ofp_header *request, uint32_t buffer_id)
4161 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4163 size_t request_len = ntohs(request->length);
4165 assert(ofconn_get_ofproto(ofconn) == ofproto);
4167 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4168 group->ofconn = ofconn;
4169 group->request = xmemdup(request, MIN(request_len, 64));
4170 group->buffer_id = buffer_id;
4175 /* Submits 'group' for processing.
4177 * If 'group' contains no operations (e.g. none were ever added, or all of the
4178 * ones that were added completed synchronously), then it is destroyed
4179 * immediately. Otherwise it is added to the ofproto's list of pending
4182 ofopgroup_submit(struct ofopgroup *group)
4184 if (!group->n_running) {
4185 ofopgroup_complete(group);
4187 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4188 group->ofproto->n_pending++;
4193 ofopgroup_complete(struct ofopgroup *group)
4195 struct ofproto *ofproto = group->ofproto;
4197 struct ofconn *abbrev_ofconn;
4198 ovs_be32 abbrev_xid;
4200 struct ofoperation *op, *next_op;
4203 assert(!group->n_running);
4206 LIST_FOR_EACH (op, group_node, &group->ops) {
4213 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4214 LIST_FOR_EACH (op, group_node, &group->ops) {
4215 if (op->type != OFOPERATION_DELETE) {
4216 struct ofpbuf *packet;
4219 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4223 error = rule_execute(op->rule, in_port, packet);
4230 if (!error && !list_is_empty(&group->ofconn_node)) {
4231 abbrev_ofconn = group->ofconn;
4232 abbrev_xid = group->request->xid;
4234 abbrev_ofconn = NULL;
4235 abbrev_xid = htonl(0);
4237 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4238 struct rule *rule = op->rule;
4240 if (!op->error && !ofproto_rule_is_hidden(rule)) {
4241 /* Check that we can just cast from ofoperation_type to
4242 * nx_flow_update_event. */
4243 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4245 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4247 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4250 ofmonitor_report(ofproto->connmgr, rule,
4251 (enum nx_flow_update_event) op->type,
4252 op->reason, abbrev_ofconn, abbrev_xid);
4255 rule->pending = NULL;
4258 case OFOPERATION_ADD:
4262 ofproto_rule_destroy__(op->victim);
4263 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4264 if (vid_mask == VLAN_VID_MASK) {
4265 if (ofproto->vlan_bitmap) {
4266 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4267 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4268 bitmap_set1(ofproto->vlan_bitmap, vid);
4269 ofproto->vlans_changed = true;
4272 ofproto->vlans_changed = true;
4276 oftable_substitute_rule(rule, op->victim);
4277 ofproto_rule_destroy__(rule);
4281 case OFOPERATION_DELETE:
4283 ofproto_rule_destroy__(rule);
4287 case OFOPERATION_MODIFY:
4289 rule->modified = time_msec();
4291 rule->flow_cookie = op->flow_cookie;
4293 free(rule->ofpacts);
4294 rule->ofpacts = op->ofpacts;
4295 rule->ofpacts_len = op->ofpacts_len;
4297 op->ofpacts_len = 0;
4306 ofoperation_destroy(op);
4309 ofmonitor_flush(ofproto->connmgr);
4311 if (!list_is_empty(&group->ofproto_node)) {
4312 assert(ofproto->n_pending > 0);
4313 ofproto->n_pending--;
4314 list_remove(&group->ofproto_node);
4316 if (!list_is_empty(&group->ofconn_node)) {
4317 list_remove(&group->ofconn_node);
4319 ofconn_send_error(group->ofconn, group->request, error);
4321 connmgr_retry(ofproto->connmgr);
4323 free(group->request);
4327 /* Initiates a new operation on 'rule', of the specified 'type', within
4328 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4330 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4331 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4333 * Returns the newly created ofoperation (which is also available as
4334 * rule->pending). */
4335 static struct ofoperation *
4336 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4337 enum ofoperation_type type,
4338 enum ofp_flow_removed_reason reason)
4340 struct ofproto *ofproto = group->ofproto;
4341 struct ofoperation *op;
4343 assert(!rule->pending);
4345 op = rule->pending = xzalloc(sizeof *op);
4347 list_push_back(&group->ops, &op->group_node);
4350 op->reason = reason;
4351 op->flow_cookie = rule->flow_cookie;
4355 if (type == OFOPERATION_DELETE) {
4356 hmap_insert(&ofproto->deletions, &op->hmap_node,
4357 cls_rule_hash(&rule->cr, rule->table_id));
4364 ofoperation_destroy(struct ofoperation *op)
4366 struct ofopgroup *group = op->group;
4369 op->rule->pending = NULL;
4371 if (op->type == OFOPERATION_DELETE) {
4372 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4374 list_remove(&op->group_node);
4379 /* Indicates that 'op' completed with status 'error', which is either 0 to
4380 * indicate success or an OpenFlow error code on failure.
4382 * If 'error' is 0, indicating success, the operation will be committed
4383 * permanently to the flow table. There is one interesting subcase:
4385 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4386 * the flow table (the "victim" rule) by a new one, then the caller must
4387 * have uninitialized any derived state in the victim rule, as in step 5 in
4388 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4389 * performs steps 6 and 7 for the victim rule, most notably by calling its
4390 * ->rule_dealloc() function.
4392 * If 'error' is nonzero, then generally the operation will be rolled back:
4394 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4395 * restores the original rule. The caller must have uninitialized any
4396 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4397 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4398 * and 7 for the new rule, calling its ->rule_dealloc() function.
4400 * - If 'op' is a "modify flow" operation, ofproto restores the original
4403 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4404 * allowed to fail. It must always succeed.
4406 * Please see the large comment in ofproto/ofproto-provider.h titled
4407 * "Asynchronous Operation Support" for more information. */
4409 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4411 struct ofopgroup *group = op->group;
4413 assert(op->rule->pending == op);
4414 assert(group->n_running > 0);
4415 assert(!error || op->type != OFOPERATION_DELETE);
4418 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4419 ofopgroup_complete(group);
4424 ofoperation_get_victim(struct ofoperation *op)
4426 assert(op->type == OFOPERATION_ADD);
4431 pick_datapath_id(const struct ofproto *ofproto)
4433 const struct ofport *port;
4435 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4437 uint8_t ea[ETH_ADDR_LEN];
4440 error = netdev_get_etheraddr(port->netdev, ea);
4442 return eth_addr_to_uint64(ea);
4444 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4445 ofproto->name, netdev_get_name(port->netdev),
4448 return ofproto->fallback_dpid;
4452 pick_fallback_dpid(void)
4454 uint8_t ea[ETH_ADDR_LEN];
4455 eth_addr_nicira_random(ea);
4456 return eth_addr_to_uint64(ea);
4459 /* Table overflow policy. */
4461 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4462 * is not configured to evict rules or if the table contains no evictable
4463 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4465 static struct rule *
4466 choose_rule_to_evict(struct oftable *table)
4468 struct eviction_group *evg;
4470 if (!table->eviction_fields) {
4474 /* In the common case, the outer and inner loops here will each be entered
4477 * - The inner loop normally "return"s in its first iteration. If the
4478 * eviction group has any evictable rules, then it always returns in
4481 * - The outer loop only iterates more than once if the largest eviction
4482 * group has no evictable rules.
4484 * - The outer loop can exit only if table's 'max_flows' is all filled up
4485 * by unevictable rules'. */
4486 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4489 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4490 if (rule->evictable) {
4499 /* Searches 'ofproto' for tables that have more flows than their configured
4500 * maximum and that have flow eviction enabled, and evicts as many flows as
4501 * necessary and currently feasible from them.
4503 * This triggers only when an OpenFlow table has N flows in it and then the
4504 * client configures a maximum number of flows less than N. */
4506 ofproto_evict(struct ofproto *ofproto)
4508 struct ofopgroup *group;
4509 struct oftable *table;
4511 group = ofopgroup_create_unattached(ofproto);
4512 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4513 while (classifier_count(&table->cls) > table->max_flows
4514 && table->eviction_fields) {
4517 rule = choose_rule_to_evict(table);
4518 if (!rule || rule->pending) {
4522 ofoperation_create(group, rule,
4523 OFOPERATION_DELETE, OFPRR_EVICTION);
4524 oftable_remove_rule(rule);
4525 ofproto->ofproto_class->rule_destruct(rule);
4528 ofopgroup_submit(group);
4531 /* Eviction groups. */
4533 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4534 * rules. The priority contains low-order random bits to ensure that eviction
4535 * groups with the same number of rules are prioritized randomly. */
4537 eviction_group_priority(size_t n_rules)
4539 uint16_t size = MIN(UINT16_MAX, n_rules);
4540 return (size << 16) | random_uint16();
4543 /* Updates 'evg', an eviction_group within 'table', following a change that
4544 * adds or removes rules in 'evg'. */
4546 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4548 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4549 eviction_group_priority(heap_count(&evg->rules)));
4552 /* Destroys 'evg', an eviction_group within 'table':
4554 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4555 * rules themselves, just removes them from the eviction group.)
4557 * - Removes 'evg' from 'table'.
4561 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4563 while (!heap_is_empty(&evg->rules)) {
4566 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4567 rule->eviction_group = NULL;
4569 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4570 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4571 heap_destroy(&evg->rules);
4575 /* Removes 'rule' from its eviction group, if any. */
4577 eviction_group_remove_rule(struct rule *rule)
4579 if (rule->eviction_group) {
4580 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4581 struct eviction_group *evg = rule->eviction_group;
4583 rule->eviction_group = NULL;
4584 heap_remove(&evg->rules, &rule->evg_node);
4585 if (heap_is_empty(&evg->rules)) {
4586 eviction_group_destroy(table, evg);
4588 eviction_group_resized(table, evg);
4593 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4594 * returns the hash value. */
4596 eviction_group_hash_rule(struct rule *rule)
4598 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4599 const struct mf_subfield *sf;
4603 hash = table->eviction_group_id_basis;
4604 miniflow_expand(&rule->cr.match.flow, &flow);
4605 for (sf = table->eviction_fields;
4606 sf < &table->eviction_fields[table->n_eviction_fields];
4609 if (mf_are_prereqs_ok(sf->field, &flow)) {
4610 union mf_value value;
4612 mf_get_value(sf->field, &flow, &value);
4614 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4616 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4617 unsigned int start = sf->ofs + sf->n_bits;
4618 bitwise_zero(&value, sf->field->n_bytes, start,
4619 sf->field->n_bytes * 8 - start);
4621 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4623 hash = hash_int(hash, 0);
4630 /* Returns an eviction group within 'table' with the given 'id', creating one
4632 static struct eviction_group *
4633 eviction_group_find(struct oftable *table, uint32_t id)
4635 struct eviction_group *evg;
4637 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4641 evg = xmalloc(sizeof *evg);
4642 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4643 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4644 eviction_group_priority(0));
4645 heap_init(&evg->rules);
4650 /* Returns an eviction priority for 'rule'. The return value should be
4651 * interpreted so that higher priorities make a rule more attractive candidates
4654 rule_eviction_priority(struct rule *rule)
4656 long long int hard_expiration;
4657 long long int idle_expiration;
4658 long long int expiration;
4659 uint32_t expiration_offset;
4661 /* Calculate time of expiration. */
4662 hard_expiration = (rule->hard_timeout
4663 ? rule->modified + rule->hard_timeout * 1000
4665 idle_expiration = (rule->idle_timeout
4666 ? rule->used + rule->idle_timeout * 1000
4668 expiration = MIN(hard_expiration, idle_expiration);
4669 if (expiration == LLONG_MAX) {
4673 /* Calculate the time of expiration as a number of (approximate) seconds
4674 * after program startup.
4676 * This should work OK for program runs that last UINT32_MAX seconds or
4677 * less. Therefore, please restart OVS at least once every 136 years. */
4678 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4680 /* Invert the expiration offset because we're using a max-heap. */
4681 return UINT32_MAX - expiration_offset;
4684 /* Adds 'rule' to an appropriate eviction group for its oftable's
4685 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4686 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4689 * The caller must ensure that 'rule' is not already in an eviction group. */
4691 eviction_group_add_rule(struct rule *rule)
4693 struct ofproto *ofproto = rule->ofproto;
4694 struct oftable *table = &ofproto->tables[rule->table_id];
4696 if (table->eviction_fields
4697 && (rule->hard_timeout || rule->idle_timeout)) {
4698 struct eviction_group *evg;
4700 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4702 rule->eviction_group = evg;
4703 heap_insert(&evg->rules, &rule->evg_node,
4704 rule_eviction_priority(rule));
4705 eviction_group_resized(table, evg);
4711 /* Initializes 'table'. */
4713 oftable_init(struct oftable *table)
4715 memset(table, 0, sizeof *table);
4716 classifier_init(&table->cls);
4717 table->max_flows = UINT_MAX;
4720 /* Destroys 'table', including its classifier and eviction groups.
4722 * The caller is responsible for freeing 'table' itself. */
4724 oftable_destroy(struct oftable *table)
4726 assert(classifier_is_empty(&table->cls));
4727 oftable_disable_eviction(table);
4728 classifier_destroy(&table->cls);
4732 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4733 * string, then 'table' will use its default name.
4735 * This only affects the name exposed for a table exposed through the OpenFlow
4736 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4738 oftable_set_name(struct oftable *table, const char *name)
4740 if (name && name[0]) {
4741 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4742 if (!table->name || strncmp(name, table->name, len)) {
4744 table->name = xmemdup0(name, len);
4752 /* oftables support a choice of two policies when adding a rule would cause the
4753 * number of flows in the table to exceed the configured maximum number: either
4754 * they can refuse to add the new flow or they can evict some existing flow.
4755 * This function configures the former policy on 'table'. */
4757 oftable_disable_eviction(struct oftable *table)
4759 if (table->eviction_fields) {
4760 struct eviction_group *evg, *next;
4762 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4763 &table->eviction_groups_by_id) {
4764 eviction_group_destroy(table, evg);
4766 hmap_destroy(&table->eviction_groups_by_id);
4767 heap_destroy(&table->eviction_groups_by_size);
4769 free(table->eviction_fields);
4770 table->eviction_fields = NULL;
4771 table->n_eviction_fields = 0;
4775 /* oftables support a choice of two policies when adding a rule would cause the
4776 * number of flows in the table to exceed the configured maximum number: either
4777 * they can refuse to add the new flow or they can evict some existing flow.
4778 * This function configures the latter policy on 'table', with fairness based
4779 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4780 * 'n_fields' as 0 disables fairness.) */
4782 oftable_enable_eviction(struct oftable *table,
4783 const struct mf_subfield *fields, size_t n_fields)
4785 struct cls_cursor cursor;
4788 if (table->eviction_fields
4789 && n_fields == table->n_eviction_fields
4791 || !memcmp(fields, table->eviction_fields,
4792 n_fields * sizeof *fields))) {
4797 oftable_disable_eviction(table);
4799 table->n_eviction_fields = n_fields;
4800 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4802 table->eviction_group_id_basis = random_uint32();
4803 hmap_init(&table->eviction_groups_by_id);
4804 heap_init(&table->eviction_groups_by_size);
4806 cls_cursor_init(&cursor, &table->cls, NULL);
4807 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4808 eviction_group_add_rule(rule);
4812 /* Removes 'rule' from the oftable that contains it. */
4814 oftable_remove_rule(struct rule *rule)
4816 struct ofproto *ofproto = rule->ofproto;
4817 struct oftable *table = &ofproto->tables[rule->table_id];
4819 classifier_remove(&table->cls, &rule->cr);
4820 eviction_group_remove_rule(rule);
4823 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4824 * oftable that has an identical cls_rule. Returns the rule that was removed,
4825 * if any, and otherwise NULL. */
4826 static struct rule *
4827 oftable_replace_rule(struct rule *rule)
4829 struct ofproto *ofproto = rule->ofproto;
4830 struct oftable *table = &ofproto->tables[rule->table_id];
4831 struct rule *victim;
4833 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4835 eviction_group_remove_rule(victim);
4837 eviction_group_add_rule(rule);
4841 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4843 oftable_substitute_rule(struct rule *old, struct rule *new)
4846 oftable_replace_rule(new);
4848 oftable_remove_rule(old);
4852 /* unixctl commands. */
4855 ofproto_lookup(const char *name)
4857 struct ofproto *ofproto;
4859 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4861 if (!strcmp(ofproto->name, name)) {
4869 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4870 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4872 struct ofproto *ofproto;
4876 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4877 ds_put_format(&results, "%s\n", ofproto->name);
4879 unixctl_command_reply(conn, ds_cstr(&results));
4880 ds_destroy(&results);
4884 ofproto_unixctl_init(void)
4886 static bool registered;
4892 unixctl_command_register("ofproto/list", "", 0, 0,
4893 ofproto_unixctl_list, NULL);
4896 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4898 * This is deprecated. It is only for compatibility with broken device drivers
4899 * in old versions of Linux that do not properly support VLANs when VLAN
4900 * devices are not used. When broken device drivers are no longer in
4901 * widespread use, we will delete these interfaces. */
4903 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4904 * (exactly) by an OpenFlow rule in 'ofproto'. */
4906 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4908 const struct oftable *oftable;
4910 free(ofproto->vlan_bitmap);
4911 ofproto->vlan_bitmap = bitmap_allocate(4096);
4912 ofproto->vlans_changed = false;
4914 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4915 const struct cls_table *table;
4917 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4918 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4919 const struct cls_rule *rule;
4921 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4922 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4923 bitmap_set1(vlan_bitmap, vid);
4924 bitmap_set1(ofproto->vlan_bitmap, vid);
4931 /* Returns true if new VLANs have come into use by the flow table since the
4932 * last call to ofproto_get_vlan_usage().
4934 * We don't track when old VLANs stop being used. */
4936 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4938 return ofproto->vlans_changed;
4941 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4942 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4943 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4944 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4945 * then the VLAN device is un-enslaved. */
4947 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4948 uint16_t realdev_ofp_port, int vid)
4950 struct ofport *ofport;
4953 assert(vlandev_ofp_port != realdev_ofp_port);
4955 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4957 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4958 ofproto->name, vlandev_ofp_port);
4962 if (!ofproto->ofproto_class->set_realdev) {
4963 if (!vlandev_ofp_port) {
4966 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4970 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4972 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4973 ofproto->name, vlandev_ofp_port,
4974 netdev_get_name(ofport->netdev), strerror(error));