2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
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 ovs_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 ovs_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 ovs_assert(!ofproto->n_tables);
473 ovs_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 ovs_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_dp_desc(struct ofproto *p, const char *dp_desc)
604 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
608 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
610 return connmgr_set_snoops(ofproto->connmgr, snoops);
614 ofproto_set_netflow(struct ofproto *ofproto,
615 const struct netflow_options *nf_options)
617 if (nf_options && sset_is_empty(&nf_options->collectors)) {
621 if (ofproto->ofproto_class->set_netflow) {
622 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
624 return nf_options ? EOPNOTSUPP : 0;
629 ofproto_set_sflow(struct ofproto *ofproto,
630 const struct ofproto_sflow_options *oso)
632 if (oso && sset_is_empty(&oso->targets)) {
636 if (ofproto->ofproto_class->set_sflow) {
637 return ofproto->ofproto_class->set_sflow(ofproto, oso);
639 return oso ? EOPNOTSUPP : 0;
643 /* Spanning Tree Protocol (STP) configuration. */
645 /* Configures STP on 'ofproto' using the settings defined in 's'. If
646 * 's' is NULL, disables STP.
648 * Returns 0 if successful, otherwise a positive errno value. */
650 ofproto_set_stp(struct ofproto *ofproto,
651 const struct ofproto_stp_settings *s)
653 return (ofproto->ofproto_class->set_stp
654 ? ofproto->ofproto_class->set_stp(ofproto, s)
658 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
659 * 'enabled' member of 's' is false, then the other members are not
662 * Returns 0 if successful, otherwise a positive errno value. */
664 ofproto_get_stp_status(struct ofproto *ofproto,
665 struct ofproto_stp_status *s)
667 return (ofproto->ofproto_class->get_stp_status
668 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
672 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
673 * in 's'. The caller is responsible for assigning STP port numbers
674 * (using the 'port_num' member in the range of 1 through 255, inclusive)
675 * and ensuring there are no duplicates. If the 's' is NULL, then STP
676 * is disabled on the port.
678 * Returns 0 if successful, otherwise a positive errno value.*/
680 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
681 const struct ofproto_port_stp_settings *s)
683 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
685 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
686 ofproto->name, ofp_port);
690 return (ofproto->ofproto_class->set_stp_port
691 ? ofproto->ofproto_class->set_stp_port(ofport, s)
695 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
696 * 's'. If the 'enabled' member in 's' is false, then the other members
697 * are not meaningful.
699 * Returns 0 if successful, otherwise a positive errno value.*/
701 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
702 struct ofproto_port_stp_status *s)
704 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
706 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
707 "port %"PRIu16, ofproto->name, ofp_port);
711 return (ofproto->ofproto_class->get_stp_port_status
712 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
716 /* Queue DSCP configuration. */
718 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
719 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
720 * to implement QoS. Instead, it is used to implement features which require
721 * knowledge of what queues exist on a port, and some basic information about
724 * Returns 0 if successful, otherwise a positive errno value. */
726 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
727 const struct ofproto_port_queue *queues,
730 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
733 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
734 ofproto->name, ofp_port);
738 return (ofproto->ofproto_class->set_queues
739 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
743 /* Connectivity Fault Management configuration. */
745 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
747 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
749 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
750 if (ofport && ofproto->ofproto_class->set_cfm) {
751 ofproto->ofproto_class->set_cfm(ofport, NULL);
755 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
756 * basic configuration from the configuration members in 'cfm', and the remote
757 * maintenance point ID from remote_mpid. Ignores the statistics members of
760 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
762 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
763 const struct cfm_settings *s)
765 struct ofport *ofport;
768 ofport = ofproto_get_port(ofproto, ofp_port);
770 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
771 ofproto->name, ofp_port);
775 /* XXX: For configuration simplicity, we only support one remote_mpid
776 * outside of the CFM module. It's not clear if this is the correct long
777 * term solution or not. */
778 error = (ofproto->ofproto_class->set_cfm
779 ? ofproto->ofproto_class->set_cfm(ofport, s)
782 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
783 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
788 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
789 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
790 * 0 if LACP partner information is not current (generally indicating a
791 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
793 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
795 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
796 return (ofport && ofproto->ofproto_class->port_is_lacp_current
797 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
803 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
804 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
805 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
806 * configuration plus, if there is more than one slave, a bonding
809 * If 'aux' is already registered then this function updates its configuration
810 * to 's'. Otherwise, this function registers a new bundle.
812 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
815 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
816 const struct ofproto_bundle_settings *s)
818 return (ofproto->ofproto_class->bundle_set
819 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
823 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
824 * If no such bundle has been registered, this has no effect. */
826 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
828 return ofproto_bundle_register(ofproto, aux, NULL);
832 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
833 * If 'aux' is already registered then this function updates its configuration
834 * to 's'. Otherwise, this function registers a new mirror. */
836 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
837 const struct ofproto_mirror_settings *s)
839 return (ofproto->ofproto_class->mirror_set
840 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
844 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
845 * If no mirror has been registered, this has no effect. */
847 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
849 return ofproto_mirror_register(ofproto, aux, NULL);
852 /* Retrieves statistics from mirror associated with client data pointer
853 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
854 * 'bytes', respectively. If a particular counters is not supported,
855 * the appropriate argument is set to UINT64_MAX. */
857 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
858 uint64_t *packets, uint64_t *bytes)
860 if (!ofproto->ofproto_class->mirror_get_stats) {
861 *packets = *bytes = UINT64_MAX;
865 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
869 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
870 * which all packets are flooded, instead of using MAC learning. If
871 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
873 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
876 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
878 return (ofproto->ofproto_class->set_flood_vlans
879 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
883 /* Returns true if 'aux' is a registered bundle that is currently in use as the
884 * output for a mirror. */
886 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
888 return (ofproto->ofproto_class->is_mirror_output_bundle
889 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
893 /* Configuration of OpenFlow tables. */
895 /* Returns the number of OpenFlow tables in 'ofproto'. */
897 ofproto_get_n_tables(const struct ofproto *ofproto)
899 return ofproto->n_tables;
902 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
903 * settings from 's'. 'table_id' must be in the range 0 through the number of
904 * OpenFlow tables in 'ofproto' minus 1, inclusive.
906 * For read-only tables, only the name may be configured. */
908 ofproto_configure_table(struct ofproto *ofproto, int table_id,
909 const struct ofproto_table_settings *s)
911 struct oftable *table;
913 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
914 table = &ofproto->tables[table_id];
916 oftable_set_name(table, s->name);
918 if (table->flags & OFTABLE_READONLY) {
923 oftable_enable_eviction(table, s->groups, s->n_groups);
925 oftable_disable_eviction(table);
928 table->max_flows = s->max_flows;
929 if (classifier_count(&table->cls) > table->max_flows
930 && table->eviction_fields) {
931 /* 'table' contains more flows than allowed. We might not be able to
932 * evict them right away because of the asynchronous nature of flow
933 * table changes. Schedule eviction for later. */
934 switch (ofproto->state) {
936 ofproto->state = S_EVICT;
940 /* We're already deleting flows, nothing more to do. */
947 ofproto_has_snoops(const struct ofproto *ofproto)
949 return connmgr_has_snoops(ofproto->connmgr);
953 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
955 connmgr_get_snoops(ofproto->connmgr, snoops);
959 ofproto_flush__(struct ofproto *ofproto)
961 struct ofopgroup *group;
962 struct oftable *table;
964 if (ofproto->ofproto_class->flush) {
965 ofproto->ofproto_class->flush(ofproto);
968 group = ofopgroup_create_unattached(ofproto);
969 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
970 struct rule *rule, *next_rule;
971 struct cls_cursor cursor;
973 if (table->flags & OFTABLE_HIDDEN) {
977 cls_cursor_init(&cursor, &table->cls, NULL);
978 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
979 if (!rule->pending) {
980 ofoperation_create(group, rule, OFOPERATION_DELETE,
982 oftable_remove_rule(rule);
983 ofproto->ofproto_class->rule_destruct(rule);
987 ofopgroup_submit(group);
991 ofproto_destroy__(struct ofproto *ofproto)
993 struct oftable *table;
995 ovs_assert(list_is_empty(&ofproto->pending));
996 ovs_assert(!ofproto->n_pending);
998 connmgr_destroy(ofproto->connmgr);
1000 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1001 free(ofproto->name);
1002 free(ofproto->type);
1003 free(ofproto->mfr_desc);
1004 free(ofproto->hw_desc);
1005 free(ofproto->sw_desc);
1006 free(ofproto->serial_desc);
1007 free(ofproto->dp_desc);
1008 hmap_destroy(&ofproto->ports);
1009 shash_destroy(&ofproto->port_by_name);
1010 bitmap_free(ofproto->ofp_port_ids);
1011 simap_destroy(&ofproto->ofp_requests);
1013 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1014 oftable_destroy(table);
1016 free(ofproto->tables);
1018 hmap_destroy(&ofproto->deletions);
1020 free(ofproto->vlan_bitmap);
1022 ofproto->ofproto_class->dealloc(ofproto);
1026 ofproto_destroy(struct ofproto *p)
1028 struct ofport *ofport, *next_ofport;
1035 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1036 ofport_destroy(ofport);
1039 p->ofproto_class->destruct(p);
1040 ofproto_destroy__(p);
1043 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1044 * kernel datapath, for example, this destroys the datapath in the kernel, and
1045 * with the netdev-based datapath, it tears down the data structures that
1046 * represent the datapath.
1048 * The datapath should not be currently open as an ofproto. */
1050 ofproto_delete(const char *name, const char *type)
1052 const struct ofproto_class *class = ofproto_class_find__(type);
1053 return (!class ? EAFNOSUPPORT
1054 : !class->del ? EACCES
1055 : class->del(type, name));
1059 process_port_change(struct ofproto *ofproto, int error, char *devname)
1061 if (error == ENOBUFS) {
1062 reinit_ports(ofproto);
1063 } else if (!error) {
1064 update_port(ofproto, devname);
1070 ofproto_type_run(const char *datapath_type)
1072 const struct ofproto_class *class;
1075 datapath_type = ofproto_normalize_type(datapath_type);
1076 class = ofproto_class_find__(datapath_type);
1078 error = class->type_run ? class->type_run(datapath_type) : 0;
1079 if (error && error != EAGAIN) {
1080 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1081 datapath_type, strerror(error));
1087 ofproto_type_run_fast(const char *datapath_type)
1089 const struct ofproto_class *class;
1092 datapath_type = ofproto_normalize_type(datapath_type);
1093 class = ofproto_class_find__(datapath_type);
1095 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1096 if (error && error != EAGAIN) {
1097 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1098 datapath_type, strerror(error));
1104 ofproto_type_wait(const char *datapath_type)
1106 const struct ofproto_class *class;
1108 datapath_type = ofproto_normalize_type(datapath_type);
1109 class = ofproto_class_find__(datapath_type);
1111 if (class->type_wait) {
1112 class->type_wait(datapath_type);
1117 ofproto_run(struct ofproto *p)
1119 struct sset changed_netdevs;
1120 const char *changed_netdev;
1121 struct ofport *ofport;
1124 error = p->ofproto_class->run(p);
1125 if (error && error != EAGAIN) {
1126 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1129 if (p->ofproto_class->port_poll) {
1132 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1133 process_port_change(p, error, devname);
1137 /* Update OpenFlow port status for any port whose netdev has changed.
1139 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1140 * destroyed, so it's not safe to update ports directly from the
1141 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1142 * need this two-phase approach. */
1143 sset_init(&changed_netdevs);
1144 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1145 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1146 if (ofport->change_seq != change_seq) {
1147 ofport->change_seq = change_seq;
1148 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1151 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1152 update_port(p, changed_netdev);
1154 sset_destroy(&changed_netdevs);
1158 connmgr_run(p->connmgr, handle_openflow);
1162 connmgr_run(p->connmgr, NULL);
1164 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1165 p->state = S_OPENFLOW;
1170 connmgr_run(p->connmgr, NULL);
1172 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1173 connmgr_flushed(p->connmgr);
1174 p->state = S_OPENFLOW;
1182 if (time_msec() >= p->next_op_report) {
1183 long long int ago = (time_msec() - p->first_op) / 1000;
1184 long long int interval = (p->last_op - p->first_op) / 1000;
1188 ds_put_format(&s, "%d flow_mods ",
1189 p->n_add + p->n_delete + p->n_modify);
1190 if (interval == ago) {
1191 ds_put_format(&s, "in the last %lld s", ago);
1192 } else if (interval) {
1193 ds_put_format(&s, "in the %lld s starting %lld s ago",
1196 ds_put_format(&s, "%lld s ago", ago);
1199 ds_put_cstr(&s, " (");
1201 ds_put_format(&s, "%d adds, ", p->n_add);
1204 ds_put_format(&s, "%d deletes, ", p->n_delete);
1207 ds_put_format(&s, "%d modifications, ", p->n_modify);
1210 ds_put_char(&s, ')');
1212 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1215 p->n_add = p->n_delete = p->n_modify = 0;
1216 p->next_op_report = LLONG_MAX;
1222 /* Performs periodic activity required by 'ofproto' that needs to be done
1223 * with the least possible latency.
1225 * It makes sense to call this function a couple of times per poll loop, to
1226 * provide a significant performance boost on some benchmarks with the
1227 * ofproto-dpif implementation. */
1229 ofproto_run_fast(struct ofproto *p)
1233 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1234 if (error && error != EAGAIN) {
1235 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1236 p->name, strerror(error));
1242 ofproto_wait(struct ofproto *p)
1244 struct ofport *ofport;
1246 p->ofproto_class->wait(p);
1247 if (p->ofproto_class->port_poll_wait) {
1248 p->ofproto_class->port_poll_wait(p);
1251 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1252 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1253 poll_immediate_wake();
1259 connmgr_wait(p->connmgr, true);
1264 connmgr_wait(p->connmgr, false);
1265 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1266 poll_immediate_wake();
1273 ofproto_is_alive(const struct ofproto *p)
1275 return connmgr_has_controllers(p->connmgr);
1278 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1279 * memory_report(). */
1281 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1283 const struct oftable *table;
1284 unsigned int n_rules;
1286 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1287 simap_increase(usage, "ops",
1288 ofproto->n_pending + hmap_count(&ofproto->deletions));
1291 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1292 n_rules += classifier_count(&table->cls);
1294 simap_increase(usage, "rules", n_rules);
1296 if (ofproto->ofproto_class->get_memory_usage) {
1297 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1300 connmgr_get_memory_usage(ofproto->connmgr, usage);
1304 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1307 connmgr_get_controller_info(ofproto->connmgr, info);
1311 ofproto_free_ofproto_controller_info(struct shash *info)
1313 connmgr_free_controller_info(info);
1316 /* Makes a deep copy of 'old' into 'port'. */
1318 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1320 port->name = xstrdup(old->name);
1321 port->type = xstrdup(old->type);
1322 port->ofp_port = old->ofp_port;
1325 /* Frees memory allocated to members of 'ofproto_port'.
1327 * Do not call this function on an ofproto_port obtained from
1328 * ofproto_port_dump_next(): that function retains ownership of the data in the
1331 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1333 free(ofproto_port->name);
1334 free(ofproto_port->type);
1337 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1339 * This function provides no status indication. An error status for the entire
1340 * dump operation is provided when it is completed by calling
1341 * ofproto_port_dump_done().
1344 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1345 const struct ofproto *ofproto)
1347 dump->ofproto = ofproto;
1348 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1352 /* Attempts to retrieve another port from 'dump', which must have been created
1353 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1354 * 'port' and returns true. On failure, returns false.
1356 * Failure might indicate an actual error or merely that the last port has been
1357 * dumped. An error status for the entire dump operation is provided when it
1358 * is completed by calling ofproto_port_dump_done().
1360 * The ofproto owns the data stored in 'port'. It will remain valid until at
1361 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1362 * ofproto_port_dump_done(). */
1364 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1365 struct ofproto_port *port)
1367 const struct ofproto *ofproto = dump->ofproto;
1373 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1376 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1382 /* Completes port table dump operation 'dump', which must have been created
1383 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1384 * error-free, otherwise a positive errno value describing the problem. */
1386 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1388 const struct ofproto *ofproto = dump->ofproto;
1390 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1393 return dump->error == EOF ? 0 : dump->error;
1396 /* Returns the type to pass to netdev_open() when a datapath of type
1397 * 'datapath_type' has a port of type 'port_type', for a few special
1398 * cases when a netdev type differs from a port type. For example, when
1399 * using the userspace datapath, a port of type "internal" needs to be
1402 * Returns either 'type' itself or a string literal, which must not be
1405 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1407 const struct ofproto_class *class;
1409 datapath_type = ofproto_normalize_type(datapath_type);
1410 class = ofproto_class_find__(datapath_type);
1415 return (class->port_open_type
1416 ? class->port_open_type(datapath_type, port_type)
1420 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1421 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1422 * the port's OpenFlow port number.
1424 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1425 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1426 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1427 * 'ofp_portp' is non-null). */
1429 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1430 uint16_t *ofp_portp)
1432 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1435 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1437 const char *netdev_name = netdev_get_name(netdev);
1439 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1440 update_port(ofproto, netdev_name);
1443 struct ofproto_port ofproto_port;
1445 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1447 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1448 ofproto_port_destroy(&ofproto_port);
1453 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1454 * initializes '*port' appropriately; on failure, returns a positive errno
1457 * The caller owns the data in 'ofproto_port' and must free it with
1458 * ofproto_port_destroy() when it is no longer needed. */
1460 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1461 struct ofproto_port *port)
1465 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1467 memset(port, 0, sizeof *port);
1472 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1473 * Returns 0 if successful, otherwise a positive errno. */
1475 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1477 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1478 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1479 struct simap_node *ofp_request_node;
1482 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1483 if (ofp_request_node) {
1484 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1487 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1488 if (!error && ofport) {
1489 /* 'name' is the netdev's name and update_port() is going to close the
1490 * netdev. Just in case update_port() refers to 'name' after it
1491 * destroys 'ofport', make a copy of it around the update_port()
1493 char *devname = xstrdup(name);
1494 update_port(ofproto, devname);
1500 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1501 * performs the 'n_actions' actions in 'actions'. The new flow will not
1504 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1505 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1506 * controllers; otherwise, it will be hidden.
1508 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1510 * This is a helper function for in-band control and fail-open. */
1512 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1513 unsigned int priority,
1514 const struct ofpact *ofpacts, size_t ofpacts_len)
1516 const struct rule *rule;
1518 rule = rule_from_cls_rule(classifier_find_match_exactly(
1519 &ofproto->tables[0].cls, match, priority));
1520 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1521 ofpacts, ofpacts_len)) {
1522 struct ofputil_flow_mod fm;
1524 memset(&fm, 0, sizeof fm);
1526 fm.priority = priority;
1527 fm.buffer_id = UINT32_MAX;
1528 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1529 fm.ofpacts_len = ofpacts_len;
1530 add_flow(ofproto, NULL, &fm, NULL);
1535 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1536 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1537 * operation cannot be initiated now but may be retried later.
1539 * This is a helper function for in-band control and fail-open. */
1541 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1543 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1546 /* Searches for a rule with matching criteria exactly equal to 'target' in
1547 * ofproto's table 0 and, if it finds one, deletes it.
1549 * This is a helper function for in-band control and fail-open. */
1551 ofproto_delete_flow(struct ofproto *ofproto,
1552 const struct match *target, unsigned int priority)
1556 rule = rule_from_cls_rule(classifier_find_match_exactly(
1557 &ofproto->tables[0].cls, target, priority));
1559 /* No such rule -> success. */
1561 } else if (rule->pending) {
1562 /* An operation on the rule is already pending -> failure.
1563 * Caller must retry later if it's important. */
1566 /* Initiate deletion -> success. */
1567 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1568 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1569 oftable_remove_rule(rule);
1570 ofproto->ofproto_class->rule_destruct(rule);
1571 ofopgroup_submit(group);
1577 /* Starts the process of deleting all of the flows from all of ofproto's flow
1578 * tables and then reintroducing the flows required by in-band control and
1579 * fail-open. The process will complete in a later call to ofproto_run(). */
1581 ofproto_flush_flows(struct ofproto *ofproto)
1583 COVERAGE_INC(ofproto_flush);
1584 ofproto->state = S_FLUSH;
1588 reinit_ports(struct ofproto *p)
1590 struct ofproto_port_dump dump;
1591 struct sset devnames;
1592 struct ofport *ofport;
1593 struct ofproto_port ofproto_port;
1594 const char *devname;
1596 COVERAGE_INC(ofproto_reinit_ports);
1598 sset_init(&devnames);
1599 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1600 sset_add(&devnames, netdev_get_name(ofport->netdev));
1602 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1603 sset_add(&devnames, ofproto_port.name);
1606 SSET_FOR_EACH (devname, &devnames) {
1607 update_port(p, devname);
1609 sset_destroy(&devnames);
1613 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1616 uint16_t end_port_no = ofproto->alloc_port_no;
1618 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1619 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1621 if (ofp_port >= ofproto->max_ports
1622 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1623 /* Search for a free OpenFlow port number. We try not to
1624 * immediately reuse them to prevent problems due to old
1627 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1628 ofproto->alloc_port_no = 0;
1630 if (!bitmap_is_set(ofproto->ofp_port_ids,
1631 ofproto->alloc_port_no)) {
1632 ofp_port = ofproto->alloc_port_no;
1635 if (ofproto->alloc_port_no == end_port_no) {
1640 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1645 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1647 if (ofp_port < ofproto->max_ports) {
1648 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1652 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1653 * pointer if the netdev cannot be opened. On success, also fills in
1655 static struct netdev *
1656 ofport_open(struct ofproto *ofproto,
1657 struct ofproto_port *ofproto_port,
1658 struct ofputil_phy_port *pp)
1660 enum netdev_flags flags;
1661 struct netdev *netdev;
1664 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1666 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1667 "cannot be opened (%s)",
1669 ofproto_port->name, ofproto_port->ofp_port,
1670 ofproto_port->name, strerror(error));
1674 if (ofproto_port->ofp_port == OFPP_NONE) {
1675 if (!strcmp(ofproto->name, ofproto_port->name)) {
1676 ofproto_port->ofp_port = OFPP_LOCAL;
1678 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1679 ofproto_port->name);
1682 pp->port_no = ofproto_port->ofp_port;
1683 netdev_get_etheraddr(netdev, pp->hw_addr);
1684 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1685 netdev_get_flags(netdev, &flags);
1686 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1687 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1688 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1689 &pp->supported, &pp->peer);
1690 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1691 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1696 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1697 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1700 ofport_equal(const struct ofputil_phy_port *a,
1701 const struct ofputil_phy_port *b)
1703 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1704 && a->state == b->state
1705 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1706 && a->curr == b->curr
1707 && a->advertised == b->advertised
1708 && a->supported == b->supported
1709 && a->peer == b->peer
1710 && a->curr_speed == b->curr_speed
1711 && a->max_speed == b->max_speed);
1714 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1715 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1716 * one with the same name or port number). */
1718 ofport_install(struct ofproto *p,
1719 struct netdev *netdev, const struct ofputil_phy_port *pp)
1721 const char *netdev_name = netdev_get_name(netdev);
1722 struct ofport *ofport;
1725 /* Create ofport. */
1726 ofport = p->ofproto_class->port_alloc();
1731 ofport->ofproto = p;
1732 ofport->netdev = netdev;
1733 ofport->change_seq = netdev_change_seq(netdev);
1735 ofport->ofp_port = pp->port_no;
1737 /* Add port to 'p'. */
1738 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1739 shash_add(&p->port_by_name, netdev_name, ofport);
1741 update_mtu(p, ofport);
1743 /* Let the ofproto_class initialize its private data. */
1744 error = p->ofproto_class->port_construct(ofport);
1748 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1752 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1753 p->name, netdev_name, strerror(error));
1755 ofport_destroy__(ofport);
1757 netdev_close(netdev);
1761 /* Removes 'ofport' from 'p' and destroys it. */
1763 ofport_remove(struct ofport *ofport)
1765 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1767 ofport_destroy(ofport);
1770 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1773 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1775 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1777 ofport_remove(port);
1781 /* Updates 'port' with new 'pp' description.
1783 * Does not handle a name or port number change. The caller must implement
1784 * such a change as a delete followed by an add. */
1786 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1788 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1789 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1790 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1791 port->pp.state = pp->state;
1792 port->pp.curr = pp->curr;
1793 port->pp.advertised = pp->advertised;
1794 port->pp.supported = pp->supported;
1795 port->pp.peer = pp->peer;
1796 port->pp.curr_speed = pp->curr_speed;
1797 port->pp.max_speed = pp->max_speed;
1799 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1802 /* Update OpenFlow 'state' in 'port' and notify controller. */
1804 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1806 if (port->pp.state != state) {
1807 port->pp.state = state;
1808 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1814 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1816 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1818 if (port->ofproto->ofproto_class->set_realdev) {
1819 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1821 if (port->ofproto->ofproto_class->set_stp_port) {
1822 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1824 if (port->ofproto->ofproto_class->set_cfm) {
1825 port->ofproto->ofproto_class->set_cfm(port, NULL);
1827 if (port->ofproto->ofproto_class->bundle_remove) {
1828 port->ofproto->ofproto_class->bundle_remove(port);
1834 ofport_destroy__(struct ofport *port)
1836 struct ofproto *ofproto = port->ofproto;
1837 const char *name = netdev_get_name(port->netdev);
1839 hmap_remove(&ofproto->ports, &port->hmap_node);
1840 shash_delete(&ofproto->port_by_name,
1841 shash_find(&ofproto->port_by_name, name));
1843 netdev_close(port->netdev);
1844 ofproto->ofproto_class->port_dealloc(port);
1848 ofport_destroy(struct ofport *port)
1851 dealloc_ofp_port(port->ofproto, port->ofp_port);
1852 port->ofproto->ofproto_class->port_destruct(port);
1853 ofport_destroy__(port);
1858 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1860 struct ofport *port;
1862 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1863 hash_int(ofp_port, 0), &ofproto->ports) {
1864 if (port->ofp_port == ofp_port) {
1872 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1874 struct ofproto *ofproto = port->ofproto;
1877 if (ofproto->ofproto_class->port_get_stats) {
1878 error = ofproto->ofproto_class->port_get_stats(port, stats);
1887 update_port(struct ofproto *ofproto, const char *name)
1889 struct ofproto_port ofproto_port;
1890 struct ofputil_phy_port pp;
1891 struct netdev *netdev;
1892 struct ofport *port;
1894 COVERAGE_INC(ofproto_update_port);
1896 /* Fetch 'name''s location and properties from the datapath. */
1897 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1898 ? ofport_open(ofproto, &ofproto_port, &pp)
1901 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1902 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1903 struct netdev *old_netdev = port->netdev;
1905 /* 'name' hasn't changed location. Any properties changed? */
1906 if (!ofport_equal(&port->pp, &pp)) {
1907 ofport_modified(port, &pp);
1910 update_mtu(ofproto, port);
1912 /* Install the newly opened netdev in case it has changed.
1913 * Don't close the old netdev yet in case port_modified has to
1914 * remove a retained reference to it.*/
1915 port->netdev = netdev;
1916 port->change_seq = netdev_change_seq(netdev);
1918 if (port->ofproto->ofproto_class->port_modified) {
1919 port->ofproto->ofproto_class->port_modified(port);
1922 netdev_close(old_netdev);
1924 /* If 'port' is nonnull then its name differs from 'name' and thus
1925 * we should delete it. If we think there's a port named 'name'
1926 * then its port number must be wrong now so delete it too. */
1928 ofport_remove(port);
1930 ofport_remove_with_name(ofproto, name);
1931 ofport_install(ofproto, netdev, &pp);
1934 /* Any port named 'name' is gone now. */
1935 ofport_remove_with_name(ofproto, name);
1937 ofproto_port_destroy(&ofproto_port);
1941 init_ports(struct ofproto *p)
1943 struct ofproto_port_dump dump;
1944 struct ofproto_port ofproto_port;
1945 struct shash_node *node, *next;
1947 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1948 const char *name = ofproto_port.name;
1950 if (shash_find(&p->port_by_name, name)) {
1951 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1954 struct ofputil_phy_port pp;
1955 struct netdev *netdev;
1957 /* Check if an OpenFlow port number had been requested. */
1958 node = shash_find(&init_ofp_ports, name);
1960 const struct iface_hint *iface_hint = node->data;
1961 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
1964 netdev = ofport_open(p, &ofproto_port, &pp);
1966 ofport_install(p, netdev, &pp);
1971 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
1972 struct iface_hint *iface_hint = node->data;
1974 if (!strcmp(iface_hint->br_name, p->name)) {
1975 free(iface_hint->br_name);
1976 free(iface_hint->br_type);
1978 shash_delete(&init_ofp_ports, node);
1985 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1986 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1988 find_min_mtu(struct ofproto *p)
1990 struct ofport *ofport;
1993 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1994 struct netdev *netdev = ofport->netdev;
1997 /* Skip any internal ports, since that's what we're trying to
1999 if (!strcmp(netdev_get_type(netdev), "internal")) {
2003 if (netdev_get_mtu(netdev, &dev_mtu)) {
2006 if (!mtu || dev_mtu < mtu) {
2011 return mtu ? mtu: ETH_PAYLOAD_MAX;
2014 /* Update MTU of all datapath devices on 'p' to the minimum of the
2015 * non-datapath ports in event of 'port' added or changed. */
2017 update_mtu(struct ofproto *p, struct ofport *port)
2019 struct ofport *ofport;
2020 struct netdev *netdev = port->netdev;
2021 int dev_mtu, old_min;
2023 if (netdev_get_mtu(netdev, &dev_mtu)) {
2027 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2028 if (dev_mtu > p->min_mtu) {
2029 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2030 dev_mtu = p->min_mtu;
2033 port->mtu = dev_mtu;
2037 /* For non-internal port find new min mtu. */
2038 old_min = p->min_mtu;
2039 port->mtu = dev_mtu;
2040 p->min_mtu = find_min_mtu(p);
2041 if (p->min_mtu == old_min) {
2045 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2046 struct netdev *netdev = ofport->netdev;
2048 if (!strcmp(netdev_get_type(netdev), "internal")) {
2049 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2050 ofport->mtu = p->min_mtu;
2057 ofproto_rule_destroy__(struct rule *rule)
2060 cls_rule_destroy(&rule->cr);
2061 free(rule->ofpacts);
2062 rule->ofproto->ofproto_class->rule_dealloc(rule);
2066 /* This function allows an ofproto implementation to destroy any rules that
2067 * remain when its ->destruct() function is called. The caller must have
2068 * already uninitialized any derived members of 'rule' (step 5 described in the
2069 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2070 * This function implements steps 6 and 7.
2072 * This function should only be called from an ofproto implementation's
2073 * ->destruct() function. It is not suitable elsewhere. */
2075 ofproto_rule_destroy(struct rule *rule)
2077 ovs_assert(!rule->pending);
2078 oftable_remove_rule(rule);
2079 ofproto_rule_destroy__(rule);
2082 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2083 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2085 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2087 return (port == OFPP_ANY
2088 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2091 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2092 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2094 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2096 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2101 case OFOPERATION_ADD:
2102 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2104 case OFOPERATION_DELETE:
2107 case OFOPERATION_MODIFY:
2108 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2114 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2115 * statistics appropriately. 'packet' must have at least sizeof(struct
2116 * ofp10_packet_in) bytes of headroom.
2118 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2119 * with statistics for 'packet' either way.
2121 * Takes ownership of 'packet'. */
2123 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2127 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2129 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2130 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2133 /* Returns true if 'rule' should be hidden from the controller.
2135 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2136 * (e.g. by in-band control) and are intentionally hidden from the
2139 ofproto_rule_is_hidden(const struct rule *rule)
2141 return rule->cr.priority > UINT16_MAX;
2144 static enum oftable_flags
2145 rule_get_flags(const struct rule *rule)
2147 return rule->ofproto->tables[rule->table_id].flags;
2151 rule_is_modifiable(const struct rule *rule)
2153 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2157 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2159 ofconn_send_reply(ofconn, make_echo_reply(oh));
2164 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2166 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2167 struct ofputil_switch_features features;
2168 struct ofport *port;
2174 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2176 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2178 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2179 * if present, are always at the end.) */
2180 n_tables = ofproto->n_tables;
2181 for (i = 0; i < ofproto->n_tables; i++) {
2182 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2188 features.datapath_id = ofproto->datapath_id;
2189 features.n_buffers = pktbuf_capacity();
2190 features.n_tables = n_tables;
2191 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2192 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2194 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2196 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2197 features.auxiliary_id = 0;
2198 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2200 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2201 ofputil_put_switch_features_port(&port->pp, b);
2204 ofconn_send_reply(ofconn, b);
2209 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2211 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2212 struct ofp_switch_config *osc;
2213 enum ofp_config_flags flags;
2217 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2218 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2219 flags = ofproto->frag_handling;
2220 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2221 if (oh->version < OFP13_VERSION
2222 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2223 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2225 osc->flags = htons(flags);
2226 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2227 ofconn_send_reply(ofconn, buf);
2233 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2235 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2236 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2237 uint16_t flags = ntohs(osc->flags);
2239 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2240 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2241 enum ofp_config_flags cur = ofproto->frag_handling;
2242 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2244 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2246 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2247 ofproto->frag_handling = next;
2249 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2251 ofputil_frag_handling_to_string(next));
2255 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2256 ofconn_set_invalid_ttl_to_controller(ofconn,
2257 (oh->version < OFP13_VERSION
2258 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2260 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2265 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2266 * error message code for the caller to propagate upward. Otherwise, returns
2269 * The log message mentions 'msg_type'. */
2271 reject_slave_controller(struct ofconn *ofconn)
2273 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2274 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2275 return OFPERR_OFPBRC_EPERM;
2282 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2284 struct ofproto *p = ofconn_get_ofproto(ofconn);
2285 struct ofputil_packet_out po;
2286 struct ofpbuf *payload;
2287 uint64_t ofpacts_stub[1024 / 8];
2288 struct ofpbuf ofpacts;
2292 COVERAGE_INC(ofproto_packet_out);
2294 error = reject_slave_controller(ofconn);
2299 /* Decode message. */
2300 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2301 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2303 goto exit_free_ofpacts;
2305 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2306 error = OFPERR_OFPBRC_BAD_PORT;
2307 goto exit_free_ofpacts;
2312 if (po.buffer_id != UINT32_MAX) {
2313 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2314 if (error || !payload) {
2315 goto exit_free_ofpacts;
2318 payload = xmalloc(sizeof *payload);
2319 ofpbuf_use_const(payload, po.packet, po.packet_len);
2322 /* Verify actions against packet, then send packet if successful. */
2323 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2324 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2326 error = p->ofproto_class->packet_out(p, payload, &flow,
2327 po.ofpacts, po.ofpacts_len);
2329 ofpbuf_delete(payload);
2332 ofpbuf_uninit(&ofpacts);
2338 update_port_config(struct ofport *port,
2339 enum ofputil_port_config config,
2340 enum ofputil_port_config mask)
2342 enum ofputil_port_config old_config = port->pp.config;
2343 enum ofputil_port_config toggle;
2345 toggle = (config ^ port->pp.config) & mask;
2346 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2347 if (config & OFPUTIL_PC_PORT_DOWN) {
2348 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2350 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2352 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2355 port->pp.config ^= toggle;
2356 if (port->pp.config != old_config) {
2357 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2362 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2364 struct ofproto *p = ofconn_get_ofproto(ofconn);
2365 struct ofputil_port_mod pm;
2366 struct ofport *port;
2369 error = reject_slave_controller(ofconn);
2374 error = ofputil_decode_port_mod(oh, &pm);
2379 port = ofproto_get_port(p, pm.port_no);
2381 return OFPERR_OFPPMFC_BAD_PORT;
2382 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2383 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2385 update_port_config(port, pm.config, pm.mask);
2387 netdev_set_advertisements(port->netdev, pm.advertise);
2394 handle_desc_stats_request(struct ofconn *ofconn,
2395 const struct ofp_header *request)
2397 static const char *default_mfr_desc = "Nicira, Inc.";
2398 static const char *default_hw_desc = "Open vSwitch";
2399 static const char *default_sw_desc = VERSION;
2400 static const char *default_serial_desc = "None";
2401 static const char *default_dp_desc = "None";
2403 struct ofproto *p = ofconn_get_ofproto(ofconn);
2404 struct ofp_desc_stats *ods;
2407 msg = ofpraw_alloc_stats_reply(request, 0);
2408 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2409 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2410 sizeof ods->mfr_desc);
2411 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2412 sizeof ods->hw_desc);
2413 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2414 sizeof ods->sw_desc);
2415 ovs_strlcpy(ods->serial_num,
2416 p->serial_desc ? p->serial_desc : default_serial_desc,
2417 sizeof ods->serial_num);
2418 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2419 sizeof ods->dp_desc);
2420 ofconn_send_reply(ofconn, msg);
2426 handle_table_stats_request(struct ofconn *ofconn,
2427 const struct ofp_header *request)
2429 struct ofproto *p = ofconn_get_ofproto(ofconn);
2430 struct ofp12_table_stats *ots;
2435 /* Set up default values.
2437 * ofp12_table_stats is used as a generic structure as
2438 * it is able to hold all the fields for ofp10_table_stats
2439 * and ofp11_table_stats (and of course itself).
2441 ots = xcalloc(p->n_tables, sizeof *ots);
2442 for (i = 0; i < p->n_tables; i++) {
2443 ots[i].table_id = i;
2444 sprintf(ots[i].name, "table%zu", i);
2445 ots[i].match = htonll(OFPXMT12_MASK);
2446 ots[i].wildcards = htonll(OFPXMT12_MASK);
2447 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2448 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2449 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2450 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2451 ots[i].metadata_match = htonll(UINT64_MAX);
2452 ots[i].metadata_write = htonll(UINT64_MAX);
2453 ots[i].instructions = htonl(OFPIT11_ALL);
2454 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2455 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2456 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2459 p->ofproto_class->get_tables(p, ots);
2461 /* Post-process the tables, dropping hidden tables. */
2462 n_tables = p->n_tables;
2463 for (i = 0; i < p->n_tables; i++) {
2464 const struct oftable *table = &p->tables[i];
2466 if (table->flags & OFTABLE_HIDDEN) {
2472 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2475 if (table->max_flows < ntohl(ots[i].max_entries)) {
2476 ots[i].max_entries = htonl(table->max_flows);
2480 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2481 ofconn_send_reply(ofconn, msg);
2489 append_port_stat(struct ofport *port, struct list *replies)
2491 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2493 /* Intentionally ignore return value, since errors will set
2494 * 'stats' to all-1s, which is correct for OpenFlow, and
2495 * netdev_get_stats() will log errors. */
2496 ofproto_port_get_stats(port, &ops.stats);
2498 ofputil_append_port_stat(replies, &ops);
2502 handle_port_stats_request(struct ofconn *ofconn,
2503 const struct ofp_header *request)
2505 struct ofproto *p = ofconn_get_ofproto(ofconn);
2506 struct ofport *port;
2507 struct list replies;
2511 error = ofputil_decode_port_stats_request(request, &port_no);
2516 ofpmp_init(&replies, request);
2517 if (port_no != OFPP_ANY) {
2518 port = ofproto_get_port(p, port_no);
2520 append_port_stat(port, &replies);
2523 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2524 append_port_stat(port, &replies);
2528 ofconn_send_replies(ofconn, &replies);
2533 handle_port_desc_stats_request(struct ofconn *ofconn,
2534 const struct ofp_header *request)
2536 struct ofproto *p = ofconn_get_ofproto(ofconn);
2537 enum ofp_version version;
2538 struct ofport *port;
2539 struct list replies;
2541 ofpmp_init(&replies, request);
2543 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2544 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2545 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2548 ofconn_send_replies(ofconn, &replies);
2553 calc_flow_duration__(long long int start, long long int now,
2554 uint32_t *sec, uint32_t *nsec)
2556 long long int msecs = now - start;
2557 *sec = msecs / 1000;
2558 *nsec = (msecs % 1000) * (1000 * 1000);
2561 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2562 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2564 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2566 return (table_id == 0xff || table_id < ofproto->n_tables
2568 : OFPERR_OFPBRC_BAD_TABLE_ID);
2572 static struct oftable *
2573 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2575 struct oftable *table;
2577 for (table = &ofproto->tables[table_id];
2578 table < &ofproto->tables[ofproto->n_tables];
2580 if (!(table->flags & OFTABLE_HIDDEN)) {
2588 static struct oftable *
2589 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2591 if (table_id == 0xff) {
2592 return next_visible_table(ofproto, 0);
2593 } else if (table_id < ofproto->n_tables) {
2594 return &ofproto->tables[table_id];
2600 static struct oftable *
2601 next_matching_table(const struct ofproto *ofproto,
2602 const struct oftable *table, uint8_t table_id)
2604 return (table_id == 0xff
2605 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2609 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2611 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2612 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2614 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2615 * only once, for that table. (This can be used to access tables marked
2618 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2619 * entered at all. (Perhaps you should have validated TABLE_ID with
2620 * check_table_id().)
2622 * All parameters are evaluated multiple times.
2624 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2625 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2627 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2629 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2630 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2631 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2634 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2635 * to 'out_port' are included.
2637 * Hidden rules are always omitted.
2639 * Returns 0 on success, otherwise an OpenFlow error code. */
2641 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2642 const struct match *match,
2643 ovs_be64 cookie, ovs_be64 cookie_mask,
2644 uint16_t out_port, struct list *rules)
2646 struct oftable *table;
2650 error = check_table_id(ofproto, table_id);
2656 cls_rule_init(&cr, match, 0);
2657 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2658 struct cls_cursor cursor;
2661 cls_cursor_init(&cursor, &table->cls, &cr);
2662 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2663 if (rule->pending) {
2664 error = OFPROTO_POSTPONE;
2667 if (!ofproto_rule_is_hidden(rule)
2668 && ofproto_rule_has_out_port(rule, out_port)
2669 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2670 list_push_back(rules, &rule->ofproto_node);
2676 cls_rule_destroy(&cr);
2680 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2681 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2682 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2685 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2686 * to 'out_port' are included.
2688 * Hidden rules are always omitted.
2690 * Returns 0 on success, otherwise an OpenFlow error code. */
2692 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2693 const struct match *match, unsigned int priority,
2694 ovs_be64 cookie, ovs_be64 cookie_mask,
2695 uint16_t out_port, struct list *rules)
2697 struct oftable *table;
2701 error = check_table_id(ofproto, table_id);
2707 cls_rule_init(&cr, match, priority);
2708 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2711 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2714 if (rule->pending) {
2715 error = OFPROTO_POSTPONE;
2718 if (!ofproto_rule_is_hidden(rule)
2719 && ofproto_rule_has_out_port(rule, out_port)
2720 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2721 list_push_back(rules, &rule->ofproto_node);
2727 cls_rule_destroy(&cr);
2731 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2732 * forced into the range of a uint16_t. */
2734 age_secs(long long int age_ms)
2736 return (age_ms < 0 ? 0
2737 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2738 : (unsigned int) age_ms / 1000);
2742 handle_flow_stats_request(struct ofconn *ofconn,
2743 const struct ofp_header *request)
2745 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2746 struct ofputil_flow_stats_request fsr;
2747 struct list replies;
2752 error = ofputil_decode_flow_stats_request(&fsr, request);
2757 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2758 fsr.cookie, fsr.cookie_mask,
2759 fsr.out_port, &rules);
2764 ofpmp_init(&replies, request);
2765 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2766 long long int now = time_msec();
2767 struct ofputil_flow_stats fs;
2769 minimatch_expand(&rule->cr.match, &fs.match);
2770 fs.priority = rule->cr.priority;
2771 fs.cookie = rule->flow_cookie;
2772 fs.table_id = rule->table_id;
2773 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2775 fs.idle_timeout = rule->idle_timeout;
2776 fs.hard_timeout = rule->hard_timeout;
2777 fs.idle_age = age_secs(now - rule->used);
2778 fs.hard_age = age_secs(now - rule->modified);
2779 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2781 fs.ofpacts = rule->ofpacts;
2782 fs.ofpacts_len = rule->ofpacts_len;
2784 if (rule->send_flow_removed) {
2785 fs.flags |= OFPFF_SEND_FLOW_REM;
2786 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2787 and OFPFF13_NO_BYT_COUNTS */
2789 ofputil_append_flow_stats_reply(&fs, &replies);
2791 ofconn_send_replies(ofconn, &replies);
2797 flow_stats_ds(struct rule *rule, struct ds *results)
2799 uint64_t packet_count, byte_count;
2801 rule->ofproto->ofproto_class->rule_get_stats(rule,
2802 &packet_count, &byte_count);
2804 if (rule->table_id != 0) {
2805 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2807 ds_put_format(results, "duration=%llds, ",
2808 (time_msec() - rule->created) / 1000);
2809 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2810 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2811 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2812 cls_rule_format(&rule->cr, results);
2813 ds_put_char(results, ',');
2814 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2815 ds_put_cstr(results, "\n");
2818 /* Adds a pretty-printed description of all flows to 'results', including
2819 * hidden flows (e.g., set up by in-band control). */
2821 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2823 struct oftable *table;
2825 OFPROTO_FOR_EACH_TABLE (table, p) {
2826 struct cls_cursor cursor;
2829 cls_cursor_init(&cursor, &table->cls, NULL);
2830 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2831 flow_stats_ds(rule, results);
2836 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2837 * '*engine_type' and '*engine_id', respectively. */
2839 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2840 uint8_t *engine_type, uint8_t *engine_id)
2842 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2845 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2846 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2847 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2848 * and -1 if CFM is not enabled on 'ofp_port'. */
2850 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2852 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2853 return (ofport && ofproto->ofproto_class->get_cfm_fault
2854 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2858 /* Checks the operational status reported by the remote CFM endpoint of
2859 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2860 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2863 ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
2865 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2866 return (ofport && ofproto->ofproto_class->get_cfm_opup
2867 ? ofproto->ofproto_class->get_cfm_opup(ofport)
2871 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2872 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2873 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2874 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2876 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2877 uint16_t ofp_port, const uint64_t **rmps,
2880 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2884 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2885 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2890 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2891 * integer value between 0 and 100 to indicate the health of the port as a
2892 * percentage which is the average of cfm health of all the remote_mpids or
2893 * returns -1 if CFM is not enabled on 'ofport'. */
2895 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2897 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2898 return (ofport && ofproto->ofproto_class->get_cfm_health
2899 ? ofproto->ofproto_class->get_cfm_health(ofport)
2904 handle_aggregate_stats_request(struct ofconn *ofconn,
2905 const struct ofp_header *oh)
2907 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2908 struct ofputil_flow_stats_request request;
2909 struct ofputil_aggregate_stats stats;
2910 bool unknown_packets, unknown_bytes;
2911 struct ofpbuf *reply;
2916 error = ofputil_decode_flow_stats_request(&request, oh);
2921 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2922 request.cookie, request.cookie_mask,
2923 request.out_port, &rules);
2928 memset(&stats, 0, sizeof stats);
2929 unknown_packets = unknown_bytes = false;
2930 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2931 uint64_t packet_count;
2932 uint64_t byte_count;
2934 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2937 if (packet_count == UINT64_MAX) {
2938 unknown_packets = true;
2940 stats.packet_count += packet_count;
2943 if (byte_count == UINT64_MAX) {
2944 unknown_bytes = true;
2946 stats.byte_count += byte_count;
2951 if (unknown_packets) {
2952 stats.packet_count = UINT64_MAX;
2954 if (unknown_bytes) {
2955 stats.byte_count = UINT64_MAX;
2958 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2959 ofconn_send_reply(ofconn, reply);
2964 struct queue_stats_cbdata {
2965 struct ofport *ofport;
2966 struct list replies;
2970 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2971 const struct netdev_queue_stats *stats)
2974 struct ofputil_queue_stats oqs = {
2975 .port_no = cbdata->ofport->pp.port_no,
2976 .queue_id = queue_id,
2979 ofputil_append_queue_stat(&cbdata->replies, &oqs);
2983 handle_queue_stats_dump_cb(uint32_t queue_id,
2984 struct netdev_queue_stats *stats,
2987 struct queue_stats_cbdata *cbdata = cbdata_;
2989 put_queue_stats(cbdata, queue_id, stats);
2993 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2994 struct queue_stats_cbdata *cbdata)
2996 cbdata->ofport = port;
2997 if (queue_id == OFPQ_ALL) {
2998 netdev_dump_queue_stats(port->netdev,
2999 handle_queue_stats_dump_cb, cbdata);
3001 struct netdev_queue_stats stats;
3003 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3004 put_queue_stats(cbdata, queue_id, &stats);
3006 return OFPERR_OFPQOFC_BAD_QUEUE;
3013 handle_queue_stats_request(struct ofconn *ofconn,
3014 const struct ofp_header *rq)
3016 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3017 struct queue_stats_cbdata cbdata;
3018 struct ofport *port;
3020 struct ofputil_queue_stats_request oqsr;
3022 COVERAGE_INC(ofproto_queue_req);
3024 ofpmp_init(&cbdata.replies, rq);
3026 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3031 if (oqsr.port_no == OFPP_ANY) {
3032 error = OFPERR_OFPQOFC_BAD_QUEUE;
3033 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3034 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3039 port = ofproto_get_port(ofproto, oqsr.port_no);
3041 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3042 : OFPERR_OFPQOFC_BAD_PORT);
3045 ofconn_send_replies(ofconn, &cbdata.replies);
3047 ofpbuf_list_delete(&cbdata.replies);
3054 is_flow_deletion_pending(const struct ofproto *ofproto,
3055 const struct cls_rule *cls_rule,
3058 if (!hmap_is_empty(&ofproto->deletions)) {
3059 struct ofoperation *op;
3061 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3062 cls_rule_hash(cls_rule, table_id),
3063 &ofproto->deletions) {
3064 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3073 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3074 * in which no matching flow already exists in the flow table.
3076 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3077 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3078 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3079 * initiated now but may be retried later.
3081 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3082 * ownership remains with the caller.
3084 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3087 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3088 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3090 struct oftable *table;
3091 struct ofopgroup *group;
3092 struct rule *victim;
3097 error = check_table_id(ofproto, fm->table_id);
3103 if (fm->table_id == 0xff) {
3105 if (ofproto->ofproto_class->rule_choose_table) {
3106 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3112 ovs_assert(table_id < ofproto->n_tables);
3113 table = &ofproto->tables[table_id];
3115 table = &ofproto->tables[0];
3117 } else if (fm->table_id < ofproto->n_tables) {
3118 table = &ofproto->tables[fm->table_id];
3120 return OFPERR_OFPBRC_BAD_TABLE_ID;
3123 if (table->flags & OFTABLE_READONLY) {
3124 return OFPERR_OFPBRC_EPERM;
3127 /* Allocate new rule and initialize classifier rule. */
3128 rule = ofproto->ofproto_class->rule_alloc();
3130 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3131 ofproto->name, strerror(error));
3134 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3136 /* Serialize against pending deletion. */
3137 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3138 cls_rule_destroy(&rule->cr);
3139 ofproto->ofproto_class->rule_dealloc(rule);
3140 return OFPROTO_POSTPONE;
3143 /* Check for overlap, if requested. */
3144 if (fm->flags & OFPFF_CHECK_OVERLAP
3145 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3146 cls_rule_destroy(&rule->cr);
3147 ofproto->ofproto_class->rule_dealloc(rule);
3148 return OFPERR_OFPFMFC_OVERLAP;
3151 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3153 rule->ofproto = ofproto;
3154 rule->pending = NULL;
3155 rule->flow_cookie = fm->new_cookie;
3156 rule->created = rule->modified = rule->used = time_msec();
3157 rule->idle_timeout = fm->idle_timeout;
3158 rule->hard_timeout = fm->hard_timeout;
3159 rule->table_id = table - ofproto->tables;
3160 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3161 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3162 and OFPFF13_NO_BYT_COUNTS */
3163 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3164 rule->ofpacts_len = fm->ofpacts_len;
3165 rule->evictable = true;
3166 rule->eviction_group = NULL;
3167 rule->monitor_flags = 0;
3168 rule->add_seqno = 0;
3169 rule->modify_seqno = 0;
3171 /* Insert new rule. */
3172 victim = oftable_replace_rule(rule);
3173 if (victim && !rule_is_modifiable(victim)) {
3174 error = OFPERR_OFPBRC_EPERM;
3175 } else if (victim && victim->pending) {
3176 error = OFPROTO_POSTPONE;
3178 struct ofoperation *op;
3181 if (classifier_count(&table->cls) > table->max_flows) {
3184 was_evictable = rule->evictable;
3185 rule->evictable = false;
3186 evict = choose_rule_to_evict(table);
3187 rule->evictable = was_evictable;
3190 error = OFPERR_OFPFMFC_TABLE_FULL;
3192 } else if (evict->pending) {
3193 error = OFPROTO_POSTPONE;
3200 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3201 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3202 op->victim = victim;
3204 error = ofproto->ofproto_class->rule_construct(rule);
3206 op->group->n_running--;
3207 ofoperation_destroy(rule->pending);
3209 delete_flow__(evict, group);
3211 ofopgroup_submit(group);
3215 /* Back out if an error occurred. */
3217 oftable_substitute_rule(rule, victim);
3218 ofproto_rule_destroy__(rule);
3223 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3225 /* Modifies the rules listed in 'rules', changing their actions to match those
3228 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3231 * Returns 0 on success, otherwise an OpenFlow error code. */
3233 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3234 const struct ofputil_flow_mod *fm,
3235 const struct ofp_header *request, struct list *rules)
3237 struct ofopgroup *group;
3241 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3242 error = OFPERR_OFPBRC_EPERM;
3243 LIST_FOR_EACH (rule, ofproto_node, rules) {
3244 struct ofoperation *op;
3245 bool actions_changed;
3246 ovs_be64 new_cookie;
3248 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3250 if (rule_is_modifiable(rule)) {
3251 /* At least one rule is modifiable, don't report EPERM error. */
3257 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3258 rule->ofpacts, rule->ofpacts_len);
3259 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3261 : rule->flow_cookie);
3262 if (!actions_changed && new_cookie == rule->flow_cookie) {
3263 /* No change at all. */
3267 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3268 rule->flow_cookie = new_cookie;
3269 if (actions_changed) {
3270 op->ofpacts = rule->ofpacts;
3271 op->ofpacts_len = rule->ofpacts_len;
3272 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3273 rule->ofpacts_len = fm->ofpacts_len;
3274 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3276 ofoperation_complete(op, 0);
3279 ofopgroup_submit(group);
3285 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3286 const struct ofputil_flow_mod *fm,
3287 const struct ofp_header *request)
3289 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3292 return add_flow(ofproto, ofconn, fm, request);
3295 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3298 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3301 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3302 const struct ofputil_flow_mod *fm,
3303 const struct ofp_header *request)
3308 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3309 fm->cookie, fm->cookie_mask,
3313 } else if (list_is_empty(&rules)) {
3314 return modify_flows_add(ofproto, ofconn, fm, request);
3316 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3320 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3323 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3326 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3327 const struct ofputil_flow_mod *fm,
3328 const struct ofp_header *request)
3333 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3334 fm->priority, fm->cookie, fm->cookie_mask,
3339 } else if (list_is_empty(&rules)) {
3340 return modify_flows_add(ofproto, ofconn, fm, request);
3342 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3343 fm, request, &rules)
3348 /* OFPFC_DELETE implementation. */
3351 delete_flow__(struct rule *rule, struct ofopgroup *group)
3353 struct ofproto *ofproto = rule->ofproto;
3355 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3357 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3358 oftable_remove_rule(rule);
3359 ofproto->ofproto_class->rule_destruct(rule);
3362 /* Deletes the rules listed in 'rules'.
3364 * Returns 0 on success, otherwise an OpenFlow error code. */
3366 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3367 const struct ofp_header *request, struct list *rules)
3369 struct rule *rule, *next;
3370 struct ofopgroup *group;
3372 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3373 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3374 delete_flow__(rule, group);
3376 ofopgroup_submit(group);
3381 /* Implements OFPFC_DELETE. */
3383 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3384 const struct ofputil_flow_mod *fm,
3385 const struct ofp_header *request)
3390 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3391 fm->cookie, fm->cookie_mask,
3392 fm->out_port, &rules);
3393 return (error ? error
3394 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3399 /* Implements OFPFC_DELETE_STRICT. */
3401 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3402 const struct ofputil_flow_mod *fm,
3403 const struct ofp_header *request)
3408 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3409 fm->priority, fm->cookie, fm->cookie_mask,
3410 fm->out_port, &rules);
3411 return (error ? error
3412 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3418 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3420 struct ofputil_flow_removed fr;
3422 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3426 minimatch_expand(&rule->cr.match, &fr.match);
3427 fr.priority = rule->cr.priority;
3428 fr.cookie = rule->flow_cookie;
3430 fr.table_id = rule->table_id;
3431 calc_flow_duration__(rule->created, time_msec(),
3432 &fr.duration_sec, &fr.duration_nsec);
3433 fr.idle_timeout = rule->idle_timeout;
3434 fr.hard_timeout = rule->hard_timeout;
3435 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3438 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3442 ofproto_rule_update_used(struct rule *rule, long long int used)
3444 if (used > rule->used) {
3445 struct eviction_group *evg = rule->eviction_group;
3449 heap_change(&evg->rules, &rule->evg_node,
3450 rule_eviction_priority(rule));
3455 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3456 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3459 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3462 * ofproto implementation ->run() functions should use this function to expire
3463 * OpenFlow flows. */
3465 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3467 struct ofproto *ofproto = rule->ofproto;
3468 struct ofopgroup *group;
3470 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3472 ofproto_rule_send_removed(rule, reason);
3474 group = ofopgroup_create_unattached(ofproto);
3475 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3476 oftable_remove_rule(rule);
3477 ofproto->ofproto_class->rule_destruct(rule);
3478 ofopgroup_submit(group);
3482 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3484 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3485 struct ofputil_flow_mod fm;
3486 uint64_t ofpacts_stub[1024 / 8];
3487 struct ofpbuf ofpacts;
3491 error = reject_slave_controller(ofconn);
3496 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3497 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3500 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3501 &fm.match.flow, ofproto->max_ports);
3504 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3507 goto exit_free_ofpacts;
3510 /* Record the operation for logging a summary report. */
3511 switch (fm.command) {
3517 case OFPFC_MODIFY_STRICT:
3518 ofproto->n_modify++;
3522 case OFPFC_DELETE_STRICT:
3523 ofproto->n_delete++;
3528 if (ofproto->next_op_report == LLONG_MAX) {
3529 ofproto->first_op = now;
3530 ofproto->next_op_report = MAX(now + 10 * 1000,
3531 ofproto->op_backoff);
3532 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3534 ofproto->last_op = now;
3537 ofpbuf_uninit(&ofpacts);
3543 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3544 const struct ofputil_flow_mod *fm,
3545 const struct ofp_header *oh)
3547 if (ofproto->n_pending >= 50) {
3548 ovs_assert(!list_is_empty(&ofproto->pending));
3549 return OFPROTO_POSTPONE;
3552 switch (fm->command) {
3554 return add_flow(ofproto, ofconn, fm, oh);
3557 return modify_flows_loose(ofproto, ofconn, fm, oh);
3559 case OFPFC_MODIFY_STRICT:
3560 return modify_flow_strict(ofproto, ofconn, fm, oh);
3563 return delete_flows_loose(ofproto, ofconn, fm, oh);
3565 case OFPFC_DELETE_STRICT:
3566 return delete_flow_strict(ofproto, ofconn, fm, oh);
3569 if (fm->command > 0xff) {
3570 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3571 "flow_mod_table_id extension is not enabled",
3574 return OFPERR_OFPFMFC_BAD_COMMAND;
3579 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3581 struct ofputil_role_request rr;
3586 error = ofputil_decode_role_message(oh, &rr);
3591 if (rr.request_current_role_only) {
3592 role = ofconn_get_role(ofconn); /* NX_ROLE_* */
3598 if (ofconn_get_role(ofconn) != role
3599 && ofconn_has_pending_opgroups(ofconn)) {
3600 return OFPROTO_POSTPONE;
3603 if (rr.have_generation_id) {
3604 if (!ofconn_set_master_election_id(ofconn, rr.generation_id)) {
3605 return OFPERR_OFPRRFC_STALE;
3609 ofconn_set_role(ofconn, role);
3612 buf = ofputil_encode_role_reply(oh, role);
3613 ofconn_send_reply(ofconn, buf);
3619 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3620 const struct ofp_header *oh)
3622 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3623 enum ofputil_protocol cur, next;
3625 cur = ofconn_get_protocol(ofconn);
3626 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3627 ofconn_set_protocol(ofconn, next);
3633 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3635 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3636 enum ofputil_protocol cur, next;
3637 enum ofputil_protocol next_base;
3639 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3641 return OFPERR_OFPBRC_EPERM;
3644 cur = ofconn_get_protocol(ofconn);
3645 next = ofputil_protocol_set_base(cur, next_base);
3646 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3647 /* Avoid sending async messages in surprising protocol. */
3648 return OFPROTO_POSTPONE;
3651 ofconn_set_protocol(ofconn, next);
3656 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3657 const struct ofp_header *oh)
3659 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3662 format = ntohl(msg->format);
3663 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3664 return OFPERR_OFPBRC_EPERM;
3667 if (format != ofconn_get_packet_in_format(ofconn)
3668 && ofconn_has_pending_opgroups(ofconn)) {
3669 /* Avoid sending async message in surprsing packet in format. */
3670 return OFPROTO_POSTPONE;
3673 ofconn_set_packet_in_format(ofconn, format);
3678 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3680 const struct nx_async_config *msg = ofpmsg_body(oh);
3681 uint32_t master[OAM_N_TYPES];
3682 uint32_t slave[OAM_N_TYPES];
3684 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3685 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3686 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3688 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3689 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3690 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3692 ofconn_set_async_config(ofconn, master, slave);
3693 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3694 !ofconn_get_miss_send_len(ofconn)) {
3695 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3702 handle_nxt_set_controller_id(struct ofconn *ofconn,
3703 const struct ofp_header *oh)
3705 const struct nx_controller_id *nci = ofpmsg_body(oh);
3707 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3708 return OFPERR_NXBRC_MUST_BE_ZERO;
3711 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3716 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3720 if (ofconn_has_pending_opgroups(ofconn)) {
3721 return OFPROTO_POSTPONE;
3724 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3725 ? OFPRAW_OFPT10_BARRIER_REPLY
3726 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3727 ofconn_send_reply(ofconn, buf);
3732 ofproto_compose_flow_refresh_update(const struct rule *rule,
3733 enum nx_flow_monitor_flags flags,
3736 struct ofoperation *op = rule->pending;
3737 struct ofputil_flow_update fu;
3740 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3741 /* We'll report the final flow when the operation completes. Reporting
3742 * it now would cause a duplicate report later. */
3746 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3747 ? NXFME_ADDED : NXFME_MODIFIED);
3749 fu.idle_timeout = rule->idle_timeout;
3750 fu.hard_timeout = rule->hard_timeout;
3751 fu.table_id = rule->table_id;
3752 fu.cookie = rule->flow_cookie;
3753 minimatch_expand(&rule->cr.match, &match);
3755 fu.priority = rule->cr.priority;
3756 if (!(flags & NXFMF_ACTIONS)) {
3760 fu.ofpacts = rule->ofpacts;
3761 fu.ofpacts_len = rule->ofpacts_len;
3763 /* An operation is in progress. Use the previous version of the flow's
3764 * actions, so that when the operation commits we report the change. */
3766 case OFOPERATION_ADD:
3767 /* We already verified that there was a victim. */
3768 fu.ofpacts = op->victim->ofpacts;
3769 fu.ofpacts_len = op->victim->ofpacts_len;
3772 case OFOPERATION_MODIFY:
3774 fu.ofpacts = op->ofpacts;
3775 fu.ofpacts_len = op->ofpacts_len;
3777 fu.ofpacts = rule->ofpacts;
3778 fu.ofpacts_len = rule->ofpacts_len;
3782 case OFOPERATION_DELETE:
3783 fu.ofpacts = rule->ofpacts;
3784 fu.ofpacts_len = rule->ofpacts_len;
3792 if (list_is_empty(msgs)) {
3793 ofputil_start_flow_update(msgs);
3795 ofputil_append_flow_update(&fu, msgs);
3799 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3803 LIST_FOR_EACH (rule, ofproto_node, rules) {
3804 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3805 rule->monitor_flags = 0;
3807 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3812 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3813 struct rule *rule, uint64_t seqno,
3816 enum nx_flow_monitor_flags update;
3818 if (ofproto_rule_is_hidden(rule)) {
3823 ? ofoperation_has_out_port(rule->pending, m->out_port)
3824 : ofproto_rule_has_out_port(rule, m->out_port))) {
3829 if (rule->add_seqno > seqno) {
3830 update = NXFMF_ADD | NXFMF_MODIFY;
3831 } else if (rule->modify_seqno > seqno) {
3832 update = NXFMF_MODIFY;
3837 if (!(m->flags & update)) {
3841 update = NXFMF_INITIAL;
3844 if (!rule->monitor_flags) {
3845 list_push_back(rules, &rule->ofproto_node);
3847 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3851 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3855 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3856 const struct ofoperation *op;
3857 const struct oftable *table;
3858 struct cls_rule target;
3860 cls_rule_init_from_minimatch(&target, &m->match, 0);
3861 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3862 struct cls_cursor cursor;
3865 cls_cursor_init(&cursor, &table->cls, &target);
3866 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3867 ovs_assert(!rule->pending); /* XXX */
3868 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3872 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3873 struct rule *rule = op->rule;
3875 if (((m->table_id == 0xff
3876 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3877 : m->table_id == rule->table_id))
3878 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3879 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3882 cls_rule_destroy(&target);
3886 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3889 if (m->flags & NXFMF_INITIAL) {
3890 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3895 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3896 uint64_t seqno, struct list *rules)
3898 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3902 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3904 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3905 struct ofmonitor **monitors;
3906 size_t n_monitors, allocated_monitors;
3907 struct list replies;
3914 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3916 n_monitors = allocated_monitors = 0;
3918 struct ofputil_flow_monitor_request request;
3919 struct ofmonitor *m;
3922 retval = ofputil_decode_flow_monitor_request(&request, &b);
3923 if (retval == EOF) {
3925 } else if (retval) {
3930 if (request.table_id != 0xff
3931 && request.table_id >= ofproto->n_tables) {
3932 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3936 error = ofmonitor_create(&request, ofconn, &m);
3941 if (n_monitors >= allocated_monitors) {
3942 monitors = x2nrealloc(monitors, &allocated_monitors,
3945 monitors[n_monitors++] = m;
3949 for (i = 0; i < n_monitors; i++) {
3950 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3953 ofpmp_init(&replies, oh);
3954 ofmonitor_compose_refresh_updates(&rules, &replies);
3955 ofconn_send_replies(ofconn, &replies);
3962 for (i = 0; i < n_monitors; i++) {
3963 ofmonitor_destroy(monitors[i]);
3970 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3972 struct ofmonitor *m;
3975 id = ofputil_decode_flow_monitor_cancel(oh);
3976 m = ofmonitor_lookup(ofconn, id);
3978 return OFPERR_NXBRC_FM_BAD_ID;
3981 ofmonitor_destroy(m);
3986 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3988 const struct ofp_header *oh = msg->data;
3992 error = ofptype_decode(&type, oh);
3998 /* OpenFlow requests. */
3999 case OFPTYPE_ECHO_REQUEST:
4000 return handle_echo_request(ofconn, oh);
4002 case OFPTYPE_FEATURES_REQUEST:
4003 return handle_features_request(ofconn, oh);
4005 case OFPTYPE_GET_CONFIG_REQUEST:
4006 return handle_get_config_request(ofconn, oh);
4008 case OFPTYPE_SET_CONFIG:
4009 return handle_set_config(ofconn, oh);
4011 case OFPTYPE_PACKET_OUT:
4012 return handle_packet_out(ofconn, oh);
4014 case OFPTYPE_PORT_MOD:
4015 return handle_port_mod(ofconn, oh);
4017 case OFPTYPE_FLOW_MOD:
4018 return handle_flow_mod(ofconn, oh);
4020 case OFPTYPE_BARRIER_REQUEST:
4021 return handle_barrier_request(ofconn, oh);
4023 case OFPTYPE_ROLE_REQUEST:
4024 return handle_role_request(ofconn, oh);
4026 /* OpenFlow replies. */
4027 case OFPTYPE_ECHO_REPLY:
4030 /* Nicira extension requests. */
4031 case OFPTYPE_FLOW_MOD_TABLE_ID:
4032 return handle_nxt_flow_mod_table_id(ofconn, oh);
4034 case OFPTYPE_SET_FLOW_FORMAT:
4035 return handle_nxt_set_flow_format(ofconn, oh);
4037 case OFPTYPE_SET_PACKET_IN_FORMAT:
4038 return handle_nxt_set_packet_in_format(ofconn, oh);
4040 case OFPTYPE_SET_CONTROLLER_ID:
4041 return handle_nxt_set_controller_id(ofconn, oh);
4043 case OFPTYPE_FLOW_AGE:
4044 /* Nothing to do. */
4047 case OFPTYPE_FLOW_MONITOR_CANCEL:
4048 return handle_flow_monitor_cancel(ofconn, oh);
4050 case OFPTYPE_SET_ASYNC_CONFIG:
4051 return handle_nxt_set_async_config(ofconn, oh);
4053 /* Statistics requests. */
4054 case OFPTYPE_DESC_STATS_REQUEST:
4055 return handle_desc_stats_request(ofconn, oh);
4057 case OFPTYPE_FLOW_STATS_REQUEST:
4058 return handle_flow_stats_request(ofconn, oh);
4060 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4061 return handle_aggregate_stats_request(ofconn, oh);
4063 case OFPTYPE_TABLE_STATS_REQUEST:
4064 return handle_table_stats_request(ofconn, oh);
4066 case OFPTYPE_PORT_STATS_REQUEST:
4067 return handle_port_stats_request(ofconn, oh);
4069 case OFPTYPE_QUEUE_STATS_REQUEST:
4070 return handle_queue_stats_request(ofconn, oh);
4072 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4073 return handle_port_desc_stats_request(ofconn, oh);
4075 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4076 return handle_flow_monitor_request(ofconn, oh);
4078 /* FIXME: Change the following once they are implemented: */
4079 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4080 case OFPTYPE_GET_ASYNC_REQUEST:
4081 case OFPTYPE_METER_MOD:
4082 case OFPTYPE_GROUP_REQUEST:
4083 case OFPTYPE_GROUP_DESC_REQUEST:
4084 case OFPTYPE_GROUP_FEATURES_REQUEST:
4085 case OFPTYPE_METER_REQUEST:
4086 case OFPTYPE_METER_CONFIG_REQUEST:
4087 case OFPTYPE_METER_FEATURES_REQUEST:
4088 case OFPTYPE_TABLE_FEATURES_REQUEST:
4089 return OFPERR_OFPBRC_BAD_TYPE;
4093 case OFPTYPE_FEATURES_REPLY:
4094 case OFPTYPE_GET_CONFIG_REPLY:
4095 case OFPTYPE_PACKET_IN:
4096 case OFPTYPE_FLOW_REMOVED:
4097 case OFPTYPE_PORT_STATUS:
4098 case OFPTYPE_BARRIER_REPLY:
4099 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4100 case OFPTYPE_DESC_STATS_REPLY:
4101 case OFPTYPE_FLOW_STATS_REPLY:
4102 case OFPTYPE_QUEUE_STATS_REPLY:
4103 case OFPTYPE_PORT_STATS_REPLY:
4104 case OFPTYPE_TABLE_STATS_REPLY:
4105 case OFPTYPE_AGGREGATE_STATS_REPLY:
4106 case OFPTYPE_PORT_DESC_STATS_REPLY:
4107 case OFPTYPE_ROLE_REPLY:
4108 case OFPTYPE_FLOW_MONITOR_PAUSED:
4109 case OFPTYPE_FLOW_MONITOR_RESUMED:
4110 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4111 case OFPTYPE_GET_ASYNC_REPLY:
4112 case OFPTYPE_GROUP_REPLY:
4113 case OFPTYPE_GROUP_DESC_REPLY:
4114 case OFPTYPE_GROUP_FEATURES_REPLY:
4115 case OFPTYPE_METER_REPLY:
4116 case OFPTYPE_METER_CONFIG_REPLY:
4117 case OFPTYPE_METER_FEATURES_REPLY:
4118 case OFPTYPE_TABLE_FEATURES_REPLY:
4120 return OFPERR_OFPBRC_BAD_TYPE;
4125 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4127 int error = handle_openflow__(ofconn, ofp_msg);
4128 if (error && error != OFPROTO_POSTPONE) {
4129 ofconn_send_error(ofconn, ofp_msg->data, error);
4131 COVERAGE_INC(ofproto_recv_openflow);
4132 return error != OFPROTO_POSTPONE;
4135 /* Asynchronous operations. */
4137 /* Creates and returns a new ofopgroup that is not associated with any
4138 * OpenFlow connection.
4140 * The caller should add operations to the returned group with
4141 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4142 static struct ofopgroup *
4143 ofopgroup_create_unattached(struct ofproto *ofproto)
4145 struct ofopgroup *group = xzalloc(sizeof *group);
4146 group->ofproto = ofproto;
4147 list_init(&group->ofproto_node);
4148 list_init(&group->ops);
4149 list_init(&group->ofconn_node);
4153 /* Creates and returns a new ofopgroup for 'ofproto'.
4155 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4156 * connection. The 'request' and 'buffer_id' arguments are ignored.
4158 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4159 * If the ofopgroup eventually fails, then the error reply will include
4160 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4161 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4163 * The caller should add operations to the returned group with
4164 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4165 static struct ofopgroup *
4166 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4167 const struct ofp_header *request, uint32_t buffer_id)
4169 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4171 size_t request_len = ntohs(request->length);
4173 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4175 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4176 group->ofconn = ofconn;
4177 group->request = xmemdup(request, MIN(request_len, 64));
4178 group->buffer_id = buffer_id;
4183 /* Submits 'group' for processing.
4185 * If 'group' contains no operations (e.g. none were ever added, or all of the
4186 * ones that were added completed synchronously), then it is destroyed
4187 * immediately. Otherwise it is added to the ofproto's list of pending
4190 ofopgroup_submit(struct ofopgroup *group)
4192 if (!group->n_running) {
4193 ofopgroup_complete(group);
4195 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4196 group->ofproto->n_pending++;
4201 ofopgroup_complete(struct ofopgroup *group)
4203 struct ofproto *ofproto = group->ofproto;
4205 struct ofconn *abbrev_ofconn;
4206 ovs_be32 abbrev_xid;
4208 struct ofoperation *op, *next_op;
4211 ovs_assert(!group->n_running);
4214 LIST_FOR_EACH (op, group_node, &group->ops) {
4221 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4222 LIST_FOR_EACH (op, group_node, &group->ops) {
4223 if (op->type != OFOPERATION_DELETE) {
4224 struct ofpbuf *packet;
4227 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4231 error = rule_execute(op->rule, in_port, packet);
4238 if (!error && !list_is_empty(&group->ofconn_node)) {
4239 abbrev_ofconn = group->ofconn;
4240 abbrev_xid = group->request->xid;
4242 abbrev_ofconn = NULL;
4243 abbrev_xid = htonl(0);
4245 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4246 struct rule *rule = op->rule;
4248 if (!op->error && !ofproto_rule_is_hidden(rule)) {
4249 /* Check that we can just cast from ofoperation_type to
4250 * nx_flow_update_event. */
4251 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4253 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4255 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4258 ofmonitor_report(ofproto->connmgr, rule,
4259 (enum nx_flow_update_event) op->type,
4260 op->reason, abbrev_ofconn, abbrev_xid);
4263 rule->pending = NULL;
4266 case OFOPERATION_ADD:
4270 ofproto_rule_destroy__(op->victim);
4271 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4272 if (vid_mask == VLAN_VID_MASK) {
4273 if (ofproto->vlan_bitmap) {
4274 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4275 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4276 bitmap_set1(ofproto->vlan_bitmap, vid);
4277 ofproto->vlans_changed = true;
4280 ofproto->vlans_changed = true;
4284 oftable_substitute_rule(rule, op->victim);
4285 ofproto_rule_destroy__(rule);
4289 case OFOPERATION_DELETE:
4290 ovs_assert(!op->error);
4291 ofproto_rule_destroy__(rule);
4295 case OFOPERATION_MODIFY:
4297 rule->modified = time_msec();
4299 rule->flow_cookie = op->flow_cookie;
4301 free(rule->ofpacts);
4302 rule->ofpacts = op->ofpacts;
4303 rule->ofpacts_len = op->ofpacts_len;
4305 op->ofpacts_len = 0;
4314 ofoperation_destroy(op);
4317 ofmonitor_flush(ofproto->connmgr);
4319 if (!list_is_empty(&group->ofproto_node)) {
4320 ovs_assert(ofproto->n_pending > 0);
4321 ofproto->n_pending--;
4322 list_remove(&group->ofproto_node);
4324 if (!list_is_empty(&group->ofconn_node)) {
4325 list_remove(&group->ofconn_node);
4327 ofconn_send_error(group->ofconn, group->request, error);
4329 connmgr_retry(ofproto->connmgr);
4331 free(group->request);
4335 /* Initiates a new operation on 'rule', of the specified 'type', within
4336 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4338 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4339 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4341 * Returns the newly created ofoperation (which is also available as
4342 * rule->pending). */
4343 static struct ofoperation *
4344 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4345 enum ofoperation_type type,
4346 enum ofp_flow_removed_reason reason)
4348 struct ofproto *ofproto = group->ofproto;
4349 struct ofoperation *op;
4351 ovs_assert(!rule->pending);
4353 op = rule->pending = xzalloc(sizeof *op);
4355 list_push_back(&group->ops, &op->group_node);
4358 op->reason = reason;
4359 op->flow_cookie = rule->flow_cookie;
4363 if (type == OFOPERATION_DELETE) {
4364 hmap_insert(&ofproto->deletions, &op->hmap_node,
4365 cls_rule_hash(&rule->cr, rule->table_id));
4372 ofoperation_destroy(struct ofoperation *op)
4374 struct ofopgroup *group = op->group;
4377 op->rule->pending = NULL;
4379 if (op->type == OFOPERATION_DELETE) {
4380 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4382 list_remove(&op->group_node);
4387 /* Indicates that 'op' completed with status 'error', which is either 0 to
4388 * indicate success or an OpenFlow error code on failure.
4390 * If 'error' is 0, indicating success, the operation will be committed
4391 * permanently to the flow table. There is one interesting subcase:
4393 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4394 * the flow table (the "victim" rule) by a new one, then the caller must
4395 * have uninitialized any derived state in the victim rule, as in step 5 in
4396 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4397 * performs steps 6 and 7 for the victim rule, most notably by calling its
4398 * ->rule_dealloc() function.
4400 * If 'error' is nonzero, then generally the operation will be rolled back:
4402 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4403 * restores the original rule. The caller must have uninitialized any
4404 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4405 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4406 * and 7 for the new rule, calling its ->rule_dealloc() function.
4408 * - If 'op' is a "modify flow" operation, ofproto restores the original
4411 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4412 * allowed to fail. It must always succeed.
4414 * Please see the large comment in ofproto/ofproto-provider.h titled
4415 * "Asynchronous Operation Support" for more information. */
4417 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4419 struct ofopgroup *group = op->group;
4421 ovs_assert(op->rule->pending == op);
4422 ovs_assert(group->n_running > 0);
4423 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4426 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4427 ofopgroup_complete(group);
4432 ofoperation_get_victim(struct ofoperation *op)
4434 ovs_assert(op->type == OFOPERATION_ADD);
4439 pick_datapath_id(const struct ofproto *ofproto)
4441 const struct ofport *port;
4443 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4445 uint8_t ea[ETH_ADDR_LEN];
4448 error = netdev_get_etheraddr(port->netdev, ea);
4450 return eth_addr_to_uint64(ea);
4452 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4453 ofproto->name, netdev_get_name(port->netdev),
4456 return ofproto->fallback_dpid;
4460 pick_fallback_dpid(void)
4462 uint8_t ea[ETH_ADDR_LEN];
4463 eth_addr_nicira_random(ea);
4464 return eth_addr_to_uint64(ea);
4467 /* Table overflow policy. */
4469 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4470 * is not configured to evict rules or if the table contains no evictable
4471 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4473 static struct rule *
4474 choose_rule_to_evict(struct oftable *table)
4476 struct eviction_group *evg;
4478 if (!table->eviction_fields) {
4482 /* In the common case, the outer and inner loops here will each be entered
4485 * - The inner loop normally "return"s in its first iteration. If the
4486 * eviction group has any evictable rules, then it always returns in
4489 * - The outer loop only iterates more than once if the largest eviction
4490 * group has no evictable rules.
4492 * - The outer loop can exit only if table's 'max_flows' is all filled up
4493 * by unevictable rules'. */
4494 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4497 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4498 if (rule->evictable) {
4507 /* Searches 'ofproto' for tables that have more flows than their configured
4508 * maximum and that have flow eviction enabled, and evicts as many flows as
4509 * necessary and currently feasible from them.
4511 * This triggers only when an OpenFlow table has N flows in it and then the
4512 * client configures a maximum number of flows less than N. */
4514 ofproto_evict(struct ofproto *ofproto)
4516 struct ofopgroup *group;
4517 struct oftable *table;
4519 group = ofopgroup_create_unattached(ofproto);
4520 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4521 while (classifier_count(&table->cls) > table->max_flows
4522 && table->eviction_fields) {
4525 rule = choose_rule_to_evict(table);
4526 if (!rule || rule->pending) {
4530 ofoperation_create(group, rule,
4531 OFOPERATION_DELETE, OFPRR_EVICTION);
4532 oftable_remove_rule(rule);
4533 ofproto->ofproto_class->rule_destruct(rule);
4536 ofopgroup_submit(group);
4539 /* Eviction groups. */
4541 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4542 * rules. The priority contains low-order random bits to ensure that eviction
4543 * groups with the same number of rules are prioritized randomly. */
4545 eviction_group_priority(size_t n_rules)
4547 uint16_t size = MIN(UINT16_MAX, n_rules);
4548 return (size << 16) | random_uint16();
4551 /* Updates 'evg', an eviction_group within 'table', following a change that
4552 * adds or removes rules in 'evg'. */
4554 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4556 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4557 eviction_group_priority(heap_count(&evg->rules)));
4560 /* Destroys 'evg', an eviction_group within 'table':
4562 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4563 * rules themselves, just removes them from the eviction group.)
4565 * - Removes 'evg' from 'table'.
4569 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4571 while (!heap_is_empty(&evg->rules)) {
4574 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4575 rule->eviction_group = NULL;
4577 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4578 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4579 heap_destroy(&evg->rules);
4583 /* Removes 'rule' from its eviction group, if any. */
4585 eviction_group_remove_rule(struct rule *rule)
4587 if (rule->eviction_group) {
4588 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4589 struct eviction_group *evg = rule->eviction_group;
4591 rule->eviction_group = NULL;
4592 heap_remove(&evg->rules, &rule->evg_node);
4593 if (heap_is_empty(&evg->rules)) {
4594 eviction_group_destroy(table, evg);
4596 eviction_group_resized(table, evg);
4601 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4602 * returns the hash value. */
4604 eviction_group_hash_rule(struct rule *rule)
4606 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4607 const struct mf_subfield *sf;
4611 hash = table->eviction_group_id_basis;
4612 miniflow_expand(&rule->cr.match.flow, &flow);
4613 for (sf = table->eviction_fields;
4614 sf < &table->eviction_fields[table->n_eviction_fields];
4617 if (mf_are_prereqs_ok(sf->field, &flow)) {
4618 union mf_value value;
4620 mf_get_value(sf->field, &flow, &value);
4622 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4624 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4625 unsigned int start = sf->ofs + sf->n_bits;
4626 bitwise_zero(&value, sf->field->n_bytes, start,
4627 sf->field->n_bytes * 8 - start);
4629 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4631 hash = hash_int(hash, 0);
4638 /* Returns an eviction group within 'table' with the given 'id', creating one
4640 static struct eviction_group *
4641 eviction_group_find(struct oftable *table, uint32_t id)
4643 struct eviction_group *evg;
4645 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4649 evg = xmalloc(sizeof *evg);
4650 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4651 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4652 eviction_group_priority(0));
4653 heap_init(&evg->rules);
4658 /* Returns an eviction priority for 'rule'. The return value should be
4659 * interpreted so that higher priorities make a rule more attractive candidates
4662 rule_eviction_priority(struct rule *rule)
4664 long long int hard_expiration;
4665 long long int idle_expiration;
4666 long long int expiration;
4667 uint32_t expiration_offset;
4669 /* Calculate time of expiration. */
4670 hard_expiration = (rule->hard_timeout
4671 ? rule->modified + rule->hard_timeout * 1000
4673 idle_expiration = (rule->idle_timeout
4674 ? rule->used + rule->idle_timeout * 1000
4676 expiration = MIN(hard_expiration, idle_expiration);
4677 if (expiration == LLONG_MAX) {
4681 /* Calculate the time of expiration as a number of (approximate) seconds
4682 * after program startup.
4684 * This should work OK for program runs that last UINT32_MAX seconds or
4685 * less. Therefore, please restart OVS at least once every 136 years. */
4686 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4688 /* Invert the expiration offset because we're using a max-heap. */
4689 return UINT32_MAX - expiration_offset;
4692 /* Adds 'rule' to an appropriate eviction group for its oftable's
4693 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4694 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4697 * The caller must ensure that 'rule' is not already in an eviction group. */
4699 eviction_group_add_rule(struct rule *rule)
4701 struct ofproto *ofproto = rule->ofproto;
4702 struct oftable *table = &ofproto->tables[rule->table_id];
4704 if (table->eviction_fields
4705 && (rule->hard_timeout || rule->idle_timeout)) {
4706 struct eviction_group *evg;
4708 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4710 rule->eviction_group = evg;
4711 heap_insert(&evg->rules, &rule->evg_node,
4712 rule_eviction_priority(rule));
4713 eviction_group_resized(table, evg);
4719 /* Initializes 'table'. */
4721 oftable_init(struct oftable *table)
4723 memset(table, 0, sizeof *table);
4724 classifier_init(&table->cls);
4725 table->max_flows = UINT_MAX;
4728 /* Destroys 'table', including its classifier and eviction groups.
4730 * The caller is responsible for freeing 'table' itself. */
4732 oftable_destroy(struct oftable *table)
4734 ovs_assert(classifier_is_empty(&table->cls));
4735 oftable_disable_eviction(table);
4736 classifier_destroy(&table->cls);
4740 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4741 * string, then 'table' will use its default name.
4743 * This only affects the name exposed for a table exposed through the OpenFlow
4744 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4746 oftable_set_name(struct oftable *table, const char *name)
4748 if (name && name[0]) {
4749 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4750 if (!table->name || strncmp(name, table->name, len)) {
4752 table->name = xmemdup0(name, len);
4760 /* oftables support a choice of two policies when adding a rule would cause the
4761 * number of flows in the table to exceed the configured maximum number: either
4762 * they can refuse to add the new flow or they can evict some existing flow.
4763 * This function configures the former policy on 'table'. */
4765 oftable_disable_eviction(struct oftable *table)
4767 if (table->eviction_fields) {
4768 struct eviction_group *evg, *next;
4770 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4771 &table->eviction_groups_by_id) {
4772 eviction_group_destroy(table, evg);
4774 hmap_destroy(&table->eviction_groups_by_id);
4775 heap_destroy(&table->eviction_groups_by_size);
4777 free(table->eviction_fields);
4778 table->eviction_fields = NULL;
4779 table->n_eviction_fields = 0;
4783 /* oftables support a choice of two policies when adding a rule would cause the
4784 * number of flows in the table to exceed the configured maximum number: either
4785 * they can refuse to add the new flow or they can evict some existing flow.
4786 * This function configures the latter policy on 'table', with fairness based
4787 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4788 * 'n_fields' as 0 disables fairness.) */
4790 oftable_enable_eviction(struct oftable *table,
4791 const struct mf_subfield *fields, size_t n_fields)
4793 struct cls_cursor cursor;
4796 if (table->eviction_fields
4797 && n_fields == table->n_eviction_fields
4799 || !memcmp(fields, table->eviction_fields,
4800 n_fields * sizeof *fields))) {
4805 oftable_disable_eviction(table);
4807 table->n_eviction_fields = n_fields;
4808 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4810 table->eviction_group_id_basis = random_uint32();
4811 hmap_init(&table->eviction_groups_by_id);
4812 heap_init(&table->eviction_groups_by_size);
4814 cls_cursor_init(&cursor, &table->cls, NULL);
4815 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4816 eviction_group_add_rule(rule);
4820 /* Removes 'rule' from the oftable that contains it. */
4822 oftable_remove_rule(struct rule *rule)
4824 struct ofproto *ofproto = rule->ofproto;
4825 struct oftable *table = &ofproto->tables[rule->table_id];
4827 classifier_remove(&table->cls, &rule->cr);
4828 eviction_group_remove_rule(rule);
4831 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4832 * oftable that has an identical cls_rule. Returns the rule that was removed,
4833 * if any, and otherwise NULL. */
4834 static struct rule *
4835 oftable_replace_rule(struct rule *rule)
4837 struct ofproto *ofproto = rule->ofproto;
4838 struct oftable *table = &ofproto->tables[rule->table_id];
4839 struct rule *victim;
4841 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4843 eviction_group_remove_rule(victim);
4845 eviction_group_add_rule(rule);
4849 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4851 oftable_substitute_rule(struct rule *old, struct rule *new)
4854 oftable_replace_rule(new);
4856 oftable_remove_rule(old);
4860 /* unixctl commands. */
4863 ofproto_lookup(const char *name)
4865 struct ofproto *ofproto;
4867 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4869 if (!strcmp(ofproto->name, name)) {
4877 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4878 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4880 struct ofproto *ofproto;
4884 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4885 ds_put_format(&results, "%s\n", ofproto->name);
4887 unixctl_command_reply(conn, ds_cstr(&results));
4888 ds_destroy(&results);
4892 ofproto_unixctl_init(void)
4894 static bool registered;
4900 unixctl_command_register("ofproto/list", "", 0, 0,
4901 ofproto_unixctl_list, NULL);
4904 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4906 * This is deprecated. It is only for compatibility with broken device drivers
4907 * in old versions of Linux that do not properly support VLANs when VLAN
4908 * devices are not used. When broken device drivers are no longer in
4909 * widespread use, we will delete these interfaces. */
4911 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4912 * (exactly) by an OpenFlow rule in 'ofproto'. */
4914 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4916 const struct oftable *oftable;
4918 free(ofproto->vlan_bitmap);
4919 ofproto->vlan_bitmap = bitmap_allocate(4096);
4920 ofproto->vlans_changed = false;
4922 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4923 const struct cls_table *table;
4925 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4926 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4927 const struct cls_rule *rule;
4929 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4930 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4931 bitmap_set1(vlan_bitmap, vid);
4932 bitmap_set1(ofproto->vlan_bitmap, vid);
4939 /* Returns true if new VLANs have come into use by the flow table since the
4940 * last call to ofproto_get_vlan_usage().
4942 * We don't track when old VLANs stop being used. */
4944 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4946 return ofproto->vlans_changed;
4949 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4950 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4951 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4952 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4953 * then the VLAN device is un-enslaved. */
4955 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4956 uint16_t realdev_ofp_port, int vid)
4958 struct ofport *ofport;
4961 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
4963 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4965 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4966 ofproto->name, vlandev_ofp_port);
4970 if (!ofproto->ofproto_class->set_realdev) {
4971 if (!vlandev_ofp_port) {
4974 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4978 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4980 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4981 ofproto->name, vlandev_ofp_port,
4982 netdev_get_name(ofport->netdev), strerror(error));