2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
208 static uint64_t pick_datapath_id(const struct ofproto *);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto *);
211 static void update_mtu(struct ofproto *, struct ofport *);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class **ofproto_classes;
218 static size_t n_ofproto_classes;
219 static size_t allocated_ofproto_classes;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
224 /* Initial mappings of port to OpenFlow number mappings. */
225 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
229 /* Must be called to initialize the ofproto library.
231 * The caller may pass in 'iface_hints', which contains an shash of
232 * "iface_hint" elements indexed by the interface's name. The provider
233 * may use these hints to describe the startup configuration in order to
234 * reinitialize its state. The caller owns the provided data, so a
235 * provider will make copies of anything required. An ofproto provider
236 * will remove any existing state that is not described by the hint, and
237 * may choose to remove it all. */
239 ofproto_init(const struct shash *iface_hints)
241 struct shash_node *node;
244 ofproto_class_register(&ofproto_dpif_class);
246 /* Make a local copy, since we don't own 'iface_hints' elements. */
247 SHASH_FOR_EACH(node, iface_hints) {
248 const struct iface_hint *orig_hint = node->data;
249 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
250 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
252 new_hint->br_name = xstrdup(orig_hint->br_name);
253 new_hint->br_type = xstrdup(br_type);
254 new_hint->ofp_port = orig_hint->ofp_port;
256 shash_add(&init_ofp_ports, node->name, new_hint);
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->init(&init_ofp_ports);
264 /* 'type' should be a normalized datapath type, as returned by
265 * ofproto_normalize_type(). Returns the corresponding ofproto_class
266 * structure, or a null pointer if there is none registered for 'type'. */
267 static const struct ofproto_class *
268 ofproto_class_find__(const char *type)
272 for (i = 0; i < n_ofproto_classes; i++) {
273 const struct ofproto_class *class = ofproto_classes[i];
278 class->enumerate_types(&types);
279 found = sset_contains(&types, type);
280 sset_destroy(&types);
286 VLOG_WARN("unknown datapath type %s", type);
290 /* Registers a new ofproto class. After successful registration, new ofprotos
291 * of that type can be created using ofproto_create(). */
293 ofproto_class_register(const struct ofproto_class *new_class)
297 for (i = 0; i < n_ofproto_classes; i++) {
298 if (ofproto_classes[i] == new_class) {
303 if (n_ofproto_classes >= allocated_ofproto_classes) {
304 ofproto_classes = x2nrealloc(ofproto_classes,
305 &allocated_ofproto_classes,
306 sizeof *ofproto_classes);
308 ofproto_classes[n_ofproto_classes++] = new_class;
312 /* Unregisters a datapath provider. 'type' must have been previously
313 * registered and not currently be in use by any ofprotos. After
314 * unregistration new datapaths of that type cannot be opened using
315 * ofproto_create(). */
317 ofproto_class_unregister(const struct ofproto_class *class)
321 for (i = 0; i < n_ofproto_classes; i++) {
322 if (ofproto_classes[i] == class) {
323 for (i++; i < n_ofproto_classes; i++) {
324 ofproto_classes[i - 1] = ofproto_classes[i];
330 VLOG_WARN("attempted to unregister an ofproto class that is not "
335 /* Clears 'types' and enumerates all registered ofproto types into it. The
336 * caller must first initialize the sset. */
338 ofproto_enumerate_types(struct sset *types)
342 for (i = 0; i < n_ofproto_classes; i++) {
343 ofproto_classes[i]->enumerate_types(types);
347 /* Returns the fully spelled out name for the given ofproto 'type'.
349 * Normalized type string can be compared with strcmp(). Unnormalized type
350 * string might be the same even if they have different spellings. */
352 ofproto_normalize_type(const char *type)
354 return type && type[0] ? type : "system";
357 /* Clears 'names' and enumerates the names of all known created ofprotos with
358 * the given 'type'. The caller must first initialize the sset. Returns 0 if
359 * successful, otherwise a positive errno value.
361 * Some kinds of datapaths might not be practically enumerable. This is not
362 * considered an error. */
364 ofproto_enumerate_names(const char *type, struct sset *names)
366 const struct ofproto_class *class = ofproto_class_find__(type);
367 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
371 ofproto_create(const char *datapath_name, const char *datapath_type,
372 struct ofproto **ofprotop)
374 const struct ofproto_class *class;
375 struct ofproto *ofproto;
381 ofproto_unixctl_init();
383 datapath_type = ofproto_normalize_type(datapath_type);
384 class = ofproto_class_find__(datapath_type);
386 VLOG_WARN("could not create datapath %s of unknown type %s",
387 datapath_name, datapath_type);
391 ofproto = class->alloc();
393 VLOG_ERR("failed to allocate datapath %s of type %s",
394 datapath_name, datapath_type);
399 memset(ofproto, 0, sizeof *ofproto);
400 ofproto->ofproto_class = class;
401 ofproto->name = xstrdup(datapath_name);
402 ofproto->type = xstrdup(datapath_type);
403 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
404 hash_string(ofproto->name, 0));
405 ofproto->datapath_id = 0;
406 ofproto_set_flow_eviction_threshold(ofproto,
407 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
408 ofproto->forward_bpdu = false;
409 ofproto->fallback_dpid = pick_fallback_dpid();
410 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
411 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
412 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
413 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
414 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
415 ofproto->frag_handling = OFPC_FRAG_NORMAL;
416 hmap_init(&ofproto->ports);
417 shash_init(&ofproto->port_by_name);
418 simap_init(&ofproto->ofp_requests);
419 ofproto->max_ports = OFPP_MAX;
420 ofproto->tables = NULL;
421 ofproto->n_tables = 0;
422 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
423 ofproto->state = S_OPENFLOW;
424 list_init(&ofproto->pending);
425 ofproto->n_pending = 0;
426 hmap_init(&ofproto->deletions);
427 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
428 ofproto->first_op = ofproto->last_op = LLONG_MIN;
429 ofproto->next_op_report = LLONG_MAX;
430 ofproto->op_backoff = LLONG_MIN;
431 ofproto->vlan_bitmap = NULL;
432 ofproto->vlans_changed = false;
433 ofproto->min_mtu = INT_MAX;
435 error = ofproto->ofproto_class->construct(ofproto);
437 VLOG_ERR("failed to open datapath %s: %s",
438 datapath_name, strerror(error));
439 ofproto_destroy__(ofproto);
443 /* The "max_ports" member should have been set by ->construct(ofproto).
444 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
445 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
446 bitmap_set1(ofproto->ofp_port_ids, 0);
448 /* Check that hidden tables, if any, are at the end. */
449 assert(ofproto->n_tables);
450 for (i = 0; i + 1 < ofproto->n_tables; i++) {
451 enum oftable_flags flags = ofproto->tables[i].flags;
452 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
454 assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
457 ofproto->datapath_id = pick_datapath_id(ofproto);
464 /* Must be called (only) by an ofproto implementation in its constructor
465 * function. See the large comment on 'construct' in struct ofproto_class for
468 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
470 struct oftable *table;
472 assert(!ofproto->n_tables);
473 assert(n_tables >= 1 && n_tables <= 255);
475 ofproto->n_tables = n_tables;
476 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
477 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
482 /* To be optionally called (only) by an ofproto implementation in its
483 * constructor function. See the large comment on 'construct' in struct
484 * ofproto_class for details.
486 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
487 * will then ensure that actions passed into the ofproto implementation will
488 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
489 * function is not called, there will be no such restriction.
491 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
492 * the 'max_ports' restriction. */
494 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
496 assert(max_ports <= OFPP_MAX);
497 ofproto->max_ports = max_ports;
501 ofproto_get_datapath_id(const struct ofproto *ofproto)
503 return ofproto->datapath_id;
507 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
509 uint64_t old_dpid = p->datapath_id;
510 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
511 if (p->datapath_id != old_dpid) {
512 /* Force all active connections to reconnect, since there is no way to
513 * notify a controller that the datapath ID has changed. */
514 ofproto_reconnect_controllers(p);
519 ofproto_set_controllers(struct ofproto *p,
520 const struct ofproto_controller *controllers,
521 size_t n_controllers, uint32_t allowed_versions)
523 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
528 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
530 connmgr_set_fail_mode(p->connmgr, fail_mode);
533 /* Drops the connections between 'ofproto' and all of its controllers, forcing
534 * them to reconnect. */
536 ofproto_reconnect_controllers(struct ofproto *ofproto)
538 connmgr_reconnect(ofproto->connmgr);
541 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
542 * in-band control should guarantee access, in the same way that in-band
543 * control guarantees access to OpenFlow controllers. */
545 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
546 const struct sockaddr_in *extras, size_t n)
548 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
551 /* Sets the OpenFlow queue used by flows set up by in-band control on
552 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
553 * flows will use the default queue. */
555 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
557 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
560 /* Sets the number of flows at which eviction from the kernel flow table
563 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
565 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
566 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
568 ofproto->flow_eviction_threshold = threshold;
572 /* If forward_bpdu is true, the NORMAL action will forward frames with
573 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
574 * the NORMAL action will drop these frames. */
576 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
578 bool old_val = ofproto->forward_bpdu;
579 ofproto->forward_bpdu = forward_bpdu;
580 if (old_val != ofproto->forward_bpdu) {
581 if (ofproto->ofproto_class->forward_bpdu_changed) {
582 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
587 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
588 * 'idle_time', in seconds. */
590 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
592 if (ofproto->ofproto_class->set_mac_idle_time) {
593 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
598 ofproto_set_desc(struct ofproto *p,
599 const char *mfr_desc, const char *hw_desc,
600 const char *sw_desc, const char *serial_desc,
603 struct ofp_desc_stats *ods;
606 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
607 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
608 p->name, sizeof ods->mfr_desc);
611 p->mfr_desc = xstrdup(mfr_desc);
614 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
615 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
616 p->name, sizeof ods->hw_desc);
619 p->hw_desc = xstrdup(hw_desc);
622 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
623 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
624 p->name, sizeof ods->sw_desc);
627 p->sw_desc = xstrdup(sw_desc);
630 if (strlen(serial_desc) >= sizeof ods->serial_num) {
631 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
632 "bytes", p->name, sizeof ods->serial_num);
634 free(p->serial_desc);
635 p->serial_desc = xstrdup(serial_desc);
638 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
639 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
640 p->name, sizeof ods->dp_desc);
643 p->dp_desc = xstrdup(dp_desc);
648 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
650 return connmgr_set_snoops(ofproto->connmgr, snoops);
654 ofproto_set_netflow(struct ofproto *ofproto,
655 const struct netflow_options *nf_options)
657 if (nf_options && sset_is_empty(&nf_options->collectors)) {
661 if (ofproto->ofproto_class->set_netflow) {
662 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
664 return nf_options ? EOPNOTSUPP : 0;
669 ofproto_set_sflow(struct ofproto *ofproto,
670 const struct ofproto_sflow_options *oso)
672 if (oso && sset_is_empty(&oso->targets)) {
676 if (ofproto->ofproto_class->set_sflow) {
677 return ofproto->ofproto_class->set_sflow(ofproto, oso);
679 return oso ? EOPNOTSUPP : 0;
683 /* Spanning Tree Protocol (STP) configuration. */
685 /* Configures STP on 'ofproto' using the settings defined in 's'. If
686 * 's' is NULL, disables STP.
688 * Returns 0 if successful, otherwise a positive errno value. */
690 ofproto_set_stp(struct ofproto *ofproto,
691 const struct ofproto_stp_settings *s)
693 return (ofproto->ofproto_class->set_stp
694 ? ofproto->ofproto_class->set_stp(ofproto, s)
698 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
699 * 'enabled' member of 's' is false, then the other members are not
702 * Returns 0 if successful, otherwise a positive errno value. */
704 ofproto_get_stp_status(struct ofproto *ofproto,
705 struct ofproto_stp_status *s)
707 return (ofproto->ofproto_class->get_stp_status
708 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
712 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
713 * in 's'. The caller is responsible for assigning STP port numbers
714 * (using the 'port_num' member in the range of 1 through 255, inclusive)
715 * and ensuring there are no duplicates. If the 's' is NULL, then STP
716 * is disabled on the port.
718 * Returns 0 if successful, otherwise a positive errno value.*/
720 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
721 const struct ofproto_port_stp_settings *s)
723 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
725 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
726 ofproto->name, ofp_port);
730 return (ofproto->ofproto_class->set_stp_port
731 ? ofproto->ofproto_class->set_stp_port(ofport, s)
735 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
736 * 's'. If the 'enabled' member in 's' is false, then the other members
737 * are not meaningful.
739 * Returns 0 if successful, otherwise a positive errno value.*/
741 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
742 struct ofproto_port_stp_status *s)
744 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
746 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
747 "port %"PRIu16, ofproto->name, ofp_port);
751 return (ofproto->ofproto_class->get_stp_port_status
752 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
756 /* Queue DSCP configuration. */
758 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
759 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
760 * to implement QoS. Instead, it is used to implement features which require
761 * knowledge of what queues exist on a port, and some basic information about
764 * Returns 0 if successful, otherwise a positive errno value. */
766 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
767 const struct ofproto_port_queue *queues,
770 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
773 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
774 ofproto->name, ofp_port);
778 return (ofproto->ofproto_class->set_queues
779 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
783 /* Connectivity Fault Management configuration. */
785 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
787 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
789 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
790 if (ofport && ofproto->ofproto_class->set_cfm) {
791 ofproto->ofproto_class->set_cfm(ofport, NULL);
795 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
796 * basic configuration from the configuration members in 'cfm', and the remote
797 * maintenance point ID from remote_mpid. Ignores the statistics members of
800 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
802 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
803 const struct cfm_settings *s)
805 struct ofport *ofport;
808 ofport = ofproto_get_port(ofproto, ofp_port);
810 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
811 ofproto->name, ofp_port);
815 /* XXX: For configuration simplicity, we only support one remote_mpid
816 * outside of the CFM module. It's not clear if this is the correct long
817 * term solution or not. */
818 error = (ofproto->ofproto_class->set_cfm
819 ? ofproto->ofproto_class->set_cfm(ofport, s)
822 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
823 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
828 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
829 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
830 * 0 if LACP partner information is not current (generally indicating a
831 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
833 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
835 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
836 return (ofport && ofproto->ofproto_class->port_is_lacp_current
837 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
843 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
844 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
845 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
846 * configuration plus, if there is more than one slave, a bonding
849 * If 'aux' is already registered then this function updates its configuration
850 * to 's'. Otherwise, this function registers a new bundle.
852 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
855 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
856 const struct ofproto_bundle_settings *s)
858 return (ofproto->ofproto_class->bundle_set
859 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
863 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
864 * If no such bundle has been registered, this has no effect. */
866 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
868 return ofproto_bundle_register(ofproto, aux, NULL);
872 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
873 * If 'aux' is already registered then this function updates its configuration
874 * to 's'. Otherwise, this function registers a new mirror. */
876 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
877 const struct ofproto_mirror_settings *s)
879 return (ofproto->ofproto_class->mirror_set
880 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
884 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
885 * If no mirror has been registered, this has no effect. */
887 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
889 return ofproto_mirror_register(ofproto, aux, NULL);
892 /* Retrieves statistics from mirror associated with client data pointer
893 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
894 * 'bytes', respectively. If a particular counters is not supported,
895 * the appropriate argument is set to UINT64_MAX. */
897 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
898 uint64_t *packets, uint64_t *bytes)
900 if (!ofproto->ofproto_class->mirror_get_stats) {
901 *packets = *bytes = UINT64_MAX;
905 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
909 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
910 * which all packets are flooded, instead of using MAC learning. If
911 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
913 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
916 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
918 return (ofproto->ofproto_class->set_flood_vlans
919 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
923 /* Returns true if 'aux' is a registered bundle that is currently in use as the
924 * output for a mirror. */
926 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
928 return (ofproto->ofproto_class->is_mirror_output_bundle
929 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
933 /* Configuration of OpenFlow tables. */
935 /* Returns the number of OpenFlow tables in 'ofproto'. */
937 ofproto_get_n_tables(const struct ofproto *ofproto)
939 return ofproto->n_tables;
942 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
943 * settings from 's'. 'table_id' must be in the range 0 through the number of
944 * OpenFlow tables in 'ofproto' minus 1, inclusive.
946 * For read-only tables, only the name may be configured. */
948 ofproto_configure_table(struct ofproto *ofproto, int table_id,
949 const struct ofproto_table_settings *s)
951 struct oftable *table;
953 assert(table_id >= 0 && table_id < ofproto->n_tables);
954 table = &ofproto->tables[table_id];
956 oftable_set_name(table, s->name);
958 if (table->flags & OFTABLE_READONLY) {
963 oftable_enable_eviction(table, s->groups, s->n_groups);
965 oftable_disable_eviction(table);
968 table->max_flows = s->max_flows;
969 if (classifier_count(&table->cls) > table->max_flows
970 && table->eviction_fields) {
971 /* 'table' contains more flows than allowed. We might not be able to
972 * evict them right away because of the asynchronous nature of flow
973 * table changes. Schedule eviction for later. */
974 switch (ofproto->state) {
976 ofproto->state = S_EVICT;
980 /* We're already deleting flows, nothing more to do. */
987 ofproto_has_snoops(const struct ofproto *ofproto)
989 return connmgr_has_snoops(ofproto->connmgr);
993 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
995 connmgr_get_snoops(ofproto->connmgr, snoops);
999 ofproto_flush__(struct ofproto *ofproto)
1001 struct ofopgroup *group;
1002 struct oftable *table;
1004 if (ofproto->ofproto_class->flush) {
1005 ofproto->ofproto_class->flush(ofproto);
1008 group = ofopgroup_create_unattached(ofproto);
1009 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1010 struct rule *rule, *next_rule;
1011 struct cls_cursor cursor;
1013 if (table->flags & OFTABLE_HIDDEN) {
1017 cls_cursor_init(&cursor, &table->cls, NULL);
1018 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1019 if (!rule->pending) {
1020 ofoperation_create(group, rule, OFOPERATION_DELETE,
1022 oftable_remove_rule(rule);
1023 ofproto->ofproto_class->rule_destruct(rule);
1027 ofopgroup_submit(group);
1031 ofproto_destroy__(struct ofproto *ofproto)
1033 struct oftable *table;
1035 assert(list_is_empty(&ofproto->pending));
1036 assert(!ofproto->n_pending);
1038 connmgr_destroy(ofproto->connmgr);
1040 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1041 free(ofproto->name);
1042 free(ofproto->type);
1043 free(ofproto->mfr_desc);
1044 free(ofproto->hw_desc);
1045 free(ofproto->sw_desc);
1046 free(ofproto->serial_desc);
1047 free(ofproto->dp_desc);
1048 hmap_destroy(&ofproto->ports);
1049 shash_destroy(&ofproto->port_by_name);
1050 bitmap_free(ofproto->ofp_port_ids);
1051 simap_destroy(&ofproto->ofp_requests);
1053 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1054 oftable_destroy(table);
1056 free(ofproto->tables);
1058 hmap_destroy(&ofproto->deletions);
1060 free(ofproto->vlan_bitmap);
1062 ofproto->ofproto_class->dealloc(ofproto);
1066 ofproto_destroy(struct ofproto *p)
1068 struct ofport *ofport, *next_ofport;
1075 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1076 ofport_destroy(ofport);
1079 p->ofproto_class->destruct(p);
1080 ofproto_destroy__(p);
1083 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1084 * kernel datapath, for example, this destroys the datapath in the kernel, and
1085 * with the netdev-based datapath, it tears down the data structures that
1086 * represent the datapath.
1088 * The datapath should not be currently open as an ofproto. */
1090 ofproto_delete(const char *name, const char *type)
1092 const struct ofproto_class *class = ofproto_class_find__(type);
1093 return (!class ? EAFNOSUPPORT
1094 : !class->del ? EACCES
1095 : class->del(type, name));
1099 process_port_change(struct ofproto *ofproto, int error, char *devname)
1101 if (error == ENOBUFS) {
1102 reinit_ports(ofproto);
1103 } else if (!error) {
1104 update_port(ofproto, devname);
1110 ofproto_type_run(const char *datapath_type)
1112 const struct ofproto_class *class;
1115 datapath_type = ofproto_normalize_type(datapath_type);
1116 class = ofproto_class_find__(datapath_type);
1118 error = class->type_run ? class->type_run(datapath_type) : 0;
1119 if (error && error != EAGAIN) {
1120 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1121 datapath_type, strerror(error));
1127 ofproto_type_run_fast(const char *datapath_type)
1129 const struct ofproto_class *class;
1132 datapath_type = ofproto_normalize_type(datapath_type);
1133 class = ofproto_class_find__(datapath_type);
1135 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1136 if (error && error != EAGAIN) {
1137 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1138 datapath_type, strerror(error));
1144 ofproto_type_wait(const char *datapath_type)
1146 const struct ofproto_class *class;
1148 datapath_type = ofproto_normalize_type(datapath_type);
1149 class = ofproto_class_find__(datapath_type);
1151 if (class->type_wait) {
1152 class->type_wait(datapath_type);
1157 ofproto_run(struct ofproto *p)
1159 struct sset changed_netdevs;
1160 const char *changed_netdev;
1161 struct ofport *ofport;
1164 error = p->ofproto_class->run(p);
1165 if (error && error != EAGAIN) {
1166 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1169 if (p->ofproto_class->port_poll) {
1172 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1173 process_port_change(p, error, devname);
1177 /* Update OpenFlow port status for any port whose netdev has changed.
1179 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1180 * destroyed, so it's not safe to update ports directly from the
1181 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1182 * need this two-phase approach. */
1183 sset_init(&changed_netdevs);
1184 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1185 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1186 if (ofport->change_seq != change_seq) {
1187 ofport->change_seq = change_seq;
1188 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1191 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1192 update_port(p, changed_netdev);
1194 sset_destroy(&changed_netdevs);
1198 connmgr_run(p->connmgr, handle_openflow);
1202 connmgr_run(p->connmgr, NULL);
1204 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1205 p->state = S_OPENFLOW;
1210 connmgr_run(p->connmgr, NULL);
1212 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1213 connmgr_flushed(p->connmgr);
1214 p->state = S_OPENFLOW;
1222 if (time_msec() >= p->next_op_report) {
1223 long long int ago = (time_msec() - p->first_op) / 1000;
1224 long long int interval = (p->last_op - p->first_op) / 1000;
1228 ds_put_format(&s, "%d flow_mods ",
1229 p->n_add + p->n_delete + p->n_modify);
1230 if (interval == ago) {
1231 ds_put_format(&s, "in the last %lld s", ago);
1232 } else if (interval) {
1233 ds_put_format(&s, "in the %lld s starting %lld s ago",
1236 ds_put_format(&s, "%lld s ago", ago);
1239 ds_put_cstr(&s, " (");
1241 ds_put_format(&s, "%d adds, ", p->n_add);
1244 ds_put_format(&s, "%d deletes, ", p->n_delete);
1247 ds_put_format(&s, "%d modifications, ", p->n_modify);
1250 ds_put_char(&s, ')');
1252 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1255 p->n_add = p->n_delete = p->n_modify = 0;
1256 p->next_op_report = LLONG_MAX;
1262 /* Performs periodic activity required by 'ofproto' that needs to be done
1263 * with the least possible latency.
1265 * It makes sense to call this function a couple of times per poll loop, to
1266 * provide a significant performance boost on some benchmarks with the
1267 * ofproto-dpif implementation. */
1269 ofproto_run_fast(struct ofproto *p)
1273 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1274 if (error && error != EAGAIN) {
1275 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1276 p->name, strerror(error));
1282 ofproto_wait(struct ofproto *p)
1284 struct ofport *ofport;
1286 p->ofproto_class->wait(p);
1287 if (p->ofproto_class->port_poll_wait) {
1288 p->ofproto_class->port_poll_wait(p);
1291 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1292 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1293 poll_immediate_wake();
1299 connmgr_wait(p->connmgr, true);
1304 connmgr_wait(p->connmgr, false);
1305 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1306 poll_immediate_wake();
1313 ofproto_is_alive(const struct ofproto *p)
1315 return connmgr_has_controllers(p->connmgr);
1318 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1319 * memory_report(). */
1321 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1323 const struct oftable *table;
1324 unsigned int n_rules;
1326 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1327 simap_increase(usage, "ops",
1328 ofproto->n_pending + hmap_count(&ofproto->deletions));
1331 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1332 n_rules += classifier_count(&table->cls);
1334 simap_increase(usage, "rules", n_rules);
1336 if (ofproto->ofproto_class->get_memory_usage) {
1337 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1340 connmgr_get_memory_usage(ofproto->connmgr, usage);
1344 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1347 connmgr_get_controller_info(ofproto->connmgr, info);
1351 ofproto_free_ofproto_controller_info(struct shash *info)
1353 connmgr_free_controller_info(info);
1356 /* Makes a deep copy of 'old' into 'port'. */
1358 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1360 port->name = xstrdup(old->name);
1361 port->type = xstrdup(old->type);
1362 port->ofp_port = old->ofp_port;
1365 /* Frees memory allocated to members of 'ofproto_port'.
1367 * Do not call this function on an ofproto_port obtained from
1368 * ofproto_port_dump_next(): that function retains ownership of the data in the
1371 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1373 free(ofproto_port->name);
1374 free(ofproto_port->type);
1377 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1379 * This function provides no status indication. An error status for the entire
1380 * dump operation is provided when it is completed by calling
1381 * ofproto_port_dump_done().
1384 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1385 const struct ofproto *ofproto)
1387 dump->ofproto = ofproto;
1388 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1392 /* Attempts to retrieve another port from 'dump', which must have been created
1393 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1394 * 'port' and returns true. On failure, returns false.
1396 * Failure might indicate an actual error or merely that the last port has been
1397 * dumped. An error status for the entire dump operation is provided when it
1398 * is completed by calling ofproto_port_dump_done().
1400 * The ofproto owns the data stored in 'port'. It will remain valid until at
1401 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1402 * ofproto_port_dump_done(). */
1404 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1405 struct ofproto_port *port)
1407 const struct ofproto *ofproto = dump->ofproto;
1413 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1416 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1422 /* Completes port table dump operation 'dump', which must have been created
1423 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1424 * error-free, otherwise a positive errno value describing the problem. */
1426 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1428 const struct ofproto *ofproto = dump->ofproto;
1430 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1433 return dump->error == EOF ? 0 : dump->error;
1436 /* Returns the type to pass to netdev_open() when a datapath of type
1437 * 'datapath_type' has a port of type 'port_type', for a few special
1438 * cases when a netdev type differs from a port type. For example, when
1439 * using the userspace datapath, a port of type "internal" needs to be
1442 * Returns either 'type' itself or a string literal, which must not be
1445 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1447 const struct ofproto_class *class;
1449 datapath_type = ofproto_normalize_type(datapath_type);
1450 class = ofproto_class_find__(datapath_type);
1455 return (class->port_open_type
1456 ? class->port_open_type(datapath_type, port_type)
1460 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1461 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1462 * the port's OpenFlow port number.
1464 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1465 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1466 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1467 * 'ofp_portp' is non-null). */
1469 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1470 uint16_t *ofp_portp)
1472 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1475 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1477 const char *netdev_name = netdev_get_name(netdev);
1479 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1480 update_port(ofproto, netdev_name);
1483 struct ofproto_port ofproto_port;
1485 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1487 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1488 ofproto_port_destroy(&ofproto_port);
1493 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1494 * initializes '*port' appropriately; on failure, returns a positive errno
1497 * The caller owns the data in 'ofproto_port' and must free it with
1498 * ofproto_port_destroy() when it is no longer needed. */
1500 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1501 struct ofproto_port *port)
1505 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1507 memset(port, 0, sizeof *port);
1512 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1513 * Returns 0 if successful, otherwise a positive errno. */
1515 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1517 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1518 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1519 struct simap_node *ofp_request_node;
1522 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1523 if (ofp_request_node) {
1524 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1527 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1528 if (!error && ofport) {
1529 /* 'name' is the netdev's name and update_port() is going to close the
1530 * netdev. Just in case update_port() refers to 'name' after it
1531 * destroys 'ofport', make a copy of it around the update_port()
1533 char *devname = xstrdup(name);
1534 update_port(ofproto, devname);
1540 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1541 * performs the 'n_actions' actions in 'actions'. The new flow will not
1544 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1545 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1546 * controllers; otherwise, it will be hidden.
1548 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1550 * This is a helper function for in-band control and fail-open. */
1552 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1553 unsigned int priority,
1554 const struct ofpact *ofpacts, size_t ofpacts_len)
1556 const struct rule *rule;
1558 rule = rule_from_cls_rule(classifier_find_match_exactly(
1559 &ofproto->tables[0].cls, match, priority));
1560 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1561 ofpacts, ofpacts_len)) {
1562 struct ofputil_flow_mod fm;
1564 memset(&fm, 0, sizeof fm);
1566 fm.priority = priority;
1567 fm.buffer_id = UINT32_MAX;
1568 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1569 fm.ofpacts_len = ofpacts_len;
1570 add_flow(ofproto, NULL, &fm, NULL);
1575 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1576 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1577 * operation cannot be initiated now but may be retried later.
1579 * This is a helper function for in-band control and fail-open. */
1581 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1583 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1586 /* Searches for a rule with matching criteria exactly equal to 'target' in
1587 * ofproto's table 0 and, if it finds one, deletes it.
1589 * This is a helper function for in-band control and fail-open. */
1591 ofproto_delete_flow(struct ofproto *ofproto,
1592 const struct match *target, unsigned int priority)
1596 rule = rule_from_cls_rule(classifier_find_match_exactly(
1597 &ofproto->tables[0].cls, target, priority));
1599 /* No such rule -> success. */
1601 } else if (rule->pending) {
1602 /* An operation on the rule is already pending -> failure.
1603 * Caller must retry later if it's important. */
1606 /* Initiate deletion -> success. */
1607 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1608 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1609 oftable_remove_rule(rule);
1610 ofproto->ofproto_class->rule_destruct(rule);
1611 ofopgroup_submit(group);
1617 /* Starts the process of deleting all of the flows from all of ofproto's flow
1618 * tables and then reintroducing the flows required by in-band control and
1619 * fail-open. The process will complete in a later call to ofproto_run(). */
1621 ofproto_flush_flows(struct ofproto *ofproto)
1623 COVERAGE_INC(ofproto_flush);
1624 ofproto->state = S_FLUSH;
1628 reinit_ports(struct ofproto *p)
1630 struct ofproto_port_dump dump;
1631 struct sset devnames;
1632 struct ofport *ofport;
1633 struct ofproto_port ofproto_port;
1634 const char *devname;
1636 COVERAGE_INC(ofproto_reinit_ports);
1638 sset_init(&devnames);
1639 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1640 sset_add(&devnames, netdev_get_name(ofport->netdev));
1642 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1643 sset_add(&devnames, ofproto_port.name);
1646 SSET_FOR_EACH (devname, &devnames) {
1647 update_port(p, devname);
1649 sset_destroy(&devnames);
1653 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1657 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1658 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1660 if (ofp_port >= ofproto->max_ports
1661 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1662 bool retry = ofproto->alloc_port_no ? true : false;
1664 /* Search for a free OpenFlow port number. We try not to
1665 * immediately reuse them to prevent problems due to old
1667 while (ofp_port >= ofproto->max_ports) {
1668 for (ofproto->alloc_port_no++;
1669 ofproto->alloc_port_no < ofproto->max_ports;
1670 ofproto->alloc_port_no++) {
1671 if (!bitmap_is_set(ofproto->ofp_port_ids,
1672 ofproto->alloc_port_no)) {
1673 ofp_port = ofproto->alloc_port_no;
1677 if (ofproto->alloc_port_no >= ofproto->max_ports) {
1679 ofproto->alloc_port_no = 0;
1688 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1693 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1695 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1698 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1699 * pointer if the netdev cannot be opened. On success, also fills in
1701 static struct netdev *
1702 ofport_open(struct ofproto *ofproto,
1703 struct ofproto_port *ofproto_port,
1704 struct ofputil_phy_port *pp)
1706 enum netdev_flags flags;
1707 struct netdev *netdev;
1710 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1712 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1713 "cannot be opened (%s)",
1715 ofproto_port->name, ofproto_port->ofp_port,
1716 ofproto_port->name, strerror(error));
1720 if (ofproto_port->ofp_port == OFPP_NONE) {
1721 if (!strcmp(ofproto->name, ofproto_port->name)) {
1722 ofproto_port->ofp_port = OFPP_LOCAL;
1724 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1725 ofproto_port->name);
1728 pp->port_no = ofproto_port->ofp_port;
1729 netdev_get_etheraddr(netdev, pp->hw_addr);
1730 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1731 netdev_get_flags(netdev, &flags);
1732 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1733 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1734 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1735 &pp->supported, &pp->peer);
1736 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1737 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1742 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1743 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1746 ofport_equal(const struct ofputil_phy_port *a,
1747 const struct ofputil_phy_port *b)
1749 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1750 && a->state == b->state
1751 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1752 && a->curr == b->curr
1753 && a->advertised == b->advertised
1754 && a->supported == b->supported
1755 && a->peer == b->peer
1756 && a->curr_speed == b->curr_speed
1757 && a->max_speed == b->max_speed);
1760 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1761 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1762 * one with the same name or port number). */
1764 ofport_install(struct ofproto *p,
1765 struct netdev *netdev, const struct ofputil_phy_port *pp)
1767 const char *netdev_name = netdev_get_name(netdev);
1768 struct ofport *ofport;
1771 /* Create ofport. */
1772 ofport = p->ofproto_class->port_alloc();
1777 ofport->ofproto = p;
1778 ofport->netdev = netdev;
1779 ofport->change_seq = netdev_change_seq(netdev);
1781 ofport->ofp_port = pp->port_no;
1783 /* Add port to 'p'. */
1784 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1785 shash_add(&p->port_by_name, netdev_name, ofport);
1787 update_mtu(p, ofport);
1789 /* Let the ofproto_class initialize its private data. */
1790 error = p->ofproto_class->port_construct(ofport);
1794 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1798 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1799 p->name, netdev_name, strerror(error));
1801 ofport_destroy__(ofport);
1803 netdev_close(netdev);
1807 /* Removes 'ofport' from 'p' and destroys it. */
1809 ofport_remove(struct ofport *ofport)
1811 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1813 ofport_destroy(ofport);
1816 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1819 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1821 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1823 ofport_remove(port);
1827 /* Updates 'port' with new 'pp' description.
1829 * Does not handle a name or port number change. The caller must implement
1830 * such a change as a delete followed by an add. */
1832 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1834 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1835 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1836 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1837 port->pp.state = pp->state;
1838 port->pp.curr = pp->curr;
1839 port->pp.advertised = pp->advertised;
1840 port->pp.supported = pp->supported;
1841 port->pp.peer = pp->peer;
1842 port->pp.curr_speed = pp->curr_speed;
1843 port->pp.max_speed = pp->max_speed;
1845 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1848 /* Update OpenFlow 'state' in 'port' and notify controller. */
1850 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1852 if (port->pp.state != state) {
1853 port->pp.state = state;
1854 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1860 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1862 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1864 if (port->ofproto->ofproto_class->set_realdev) {
1865 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1867 if (port->ofproto->ofproto_class->set_stp_port) {
1868 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1870 if (port->ofproto->ofproto_class->set_cfm) {
1871 port->ofproto->ofproto_class->set_cfm(port, NULL);
1873 if (port->ofproto->ofproto_class->bundle_remove) {
1874 port->ofproto->ofproto_class->bundle_remove(port);
1880 ofport_destroy__(struct ofport *port)
1882 struct ofproto *ofproto = port->ofproto;
1883 const char *name = netdev_get_name(port->netdev);
1885 hmap_remove(&ofproto->ports, &port->hmap_node);
1886 shash_delete(&ofproto->port_by_name,
1887 shash_find(&ofproto->port_by_name, name));
1889 netdev_close(port->netdev);
1890 ofproto->ofproto_class->port_dealloc(port);
1894 ofport_destroy(struct ofport *port)
1897 dealloc_ofp_port(port->ofproto, port->ofp_port);
1898 port->ofproto->ofproto_class->port_destruct(port);
1899 ofport_destroy__(port);
1904 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1906 struct ofport *port;
1908 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1909 hash_int(ofp_port, 0), &ofproto->ports) {
1910 if (port->ofp_port == ofp_port) {
1918 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1920 struct ofproto *ofproto = port->ofproto;
1923 if (ofproto->ofproto_class->port_get_stats) {
1924 error = ofproto->ofproto_class->port_get_stats(port, stats);
1933 update_port(struct ofproto *ofproto, const char *name)
1935 struct ofproto_port ofproto_port;
1936 struct ofputil_phy_port pp;
1937 struct netdev *netdev;
1938 struct ofport *port;
1940 COVERAGE_INC(ofproto_update_port);
1942 /* Fetch 'name''s location and properties from the datapath. */
1943 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1944 ? ofport_open(ofproto, &ofproto_port, &pp)
1947 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1948 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1949 struct netdev *old_netdev = port->netdev;
1951 /* 'name' hasn't changed location. Any properties changed? */
1952 if (!ofport_equal(&port->pp, &pp)) {
1953 ofport_modified(port, &pp);
1956 update_mtu(ofproto, port);
1958 /* Install the newly opened netdev in case it has changed.
1959 * Don't close the old netdev yet in case port_modified has to
1960 * remove a retained reference to it.*/
1961 port->netdev = netdev;
1962 port->change_seq = netdev_change_seq(netdev);
1964 if (port->ofproto->ofproto_class->port_modified) {
1965 port->ofproto->ofproto_class->port_modified(port);
1968 netdev_close(old_netdev);
1970 /* If 'port' is nonnull then its name differs from 'name' and thus
1971 * we should delete it. If we think there's a port named 'name'
1972 * then its port number must be wrong now so delete it too. */
1974 ofport_remove(port);
1976 ofport_remove_with_name(ofproto, name);
1977 ofport_install(ofproto, netdev, &pp);
1980 /* Any port named 'name' is gone now. */
1981 ofport_remove_with_name(ofproto, name);
1983 ofproto_port_destroy(&ofproto_port);
1987 init_ports(struct ofproto *p)
1989 struct ofproto_port_dump dump;
1990 struct ofproto_port ofproto_port;
1991 struct shash_node *node, *next;
1993 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1994 const char *name = ofproto_port.name;
1996 if (shash_find(&p->port_by_name, name)) {
1997 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2000 struct ofputil_phy_port pp;
2001 struct netdev *netdev;
2003 /* Check if an OpenFlow port number had been requested. */
2004 node = shash_find(&init_ofp_ports, name);
2006 const struct iface_hint *iface_hint = node->data;
2007 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
2010 netdev = ofport_open(p, &ofproto_port, &pp);
2012 ofport_install(p, netdev, &pp);
2017 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2018 const struct iface_hint *iface_hint = node->data;
2020 if (!strcmp(iface_hint->br_name, p->name)) {
2021 free(iface_hint->br_name);
2022 free(iface_hint->br_type);
2023 shash_delete(&init_ofp_ports, node);
2030 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2031 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2033 find_min_mtu(struct ofproto *p)
2035 struct ofport *ofport;
2038 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2039 struct netdev *netdev = ofport->netdev;
2042 /* Skip any internal ports, since that's what we're trying to
2044 if (!strcmp(netdev_get_type(netdev), "internal")) {
2048 if (netdev_get_mtu(netdev, &dev_mtu)) {
2051 if (!mtu || dev_mtu < mtu) {
2056 return mtu ? mtu: ETH_PAYLOAD_MAX;
2059 /* Update MTU of all datapath devices on 'p' to the minimum of the
2060 * non-datapath ports in event of 'port' added or changed. */
2062 update_mtu(struct ofproto *p, struct ofport *port)
2064 struct ofport *ofport;
2065 struct netdev *netdev = port->netdev;
2066 int dev_mtu, old_min;
2068 if (netdev_get_mtu(netdev, &dev_mtu)) {
2072 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2073 if (dev_mtu > p->min_mtu) {
2074 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2075 dev_mtu = p->min_mtu;
2078 port->mtu = dev_mtu;
2082 /* For non-internal port find new min mtu. */
2083 old_min = p->min_mtu;
2084 port->mtu = dev_mtu;
2085 p->min_mtu = find_min_mtu(p);
2086 if (p->min_mtu == old_min) {
2090 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2091 struct netdev *netdev = ofport->netdev;
2093 if (!strcmp(netdev_get_type(netdev), "internal")) {
2094 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2095 ofport->mtu = p->min_mtu;
2102 ofproto_rule_destroy__(struct rule *rule)
2105 cls_rule_destroy(&rule->cr);
2106 free(rule->ofpacts);
2107 rule->ofproto->ofproto_class->rule_dealloc(rule);
2111 /* This function allows an ofproto implementation to destroy any rules that
2112 * remain when its ->destruct() function is called. The caller must have
2113 * already uninitialized any derived members of 'rule' (step 5 described in the
2114 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2115 * This function implements steps 6 and 7.
2117 * This function should only be called from an ofproto implementation's
2118 * ->destruct() function. It is not suitable elsewhere. */
2120 ofproto_rule_destroy(struct rule *rule)
2122 assert(!rule->pending);
2123 oftable_remove_rule(rule);
2124 ofproto_rule_destroy__(rule);
2127 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2128 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2130 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2132 return (port == OFPP_ANY
2133 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2136 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2137 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2139 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2141 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2146 case OFOPERATION_ADD:
2147 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2149 case OFOPERATION_DELETE:
2152 case OFOPERATION_MODIFY:
2153 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2159 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2160 * statistics appropriately. 'packet' must have at least sizeof(struct
2161 * ofp10_packet_in) bytes of headroom.
2163 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2164 * with statistics for 'packet' either way.
2166 * Takes ownership of 'packet'. */
2168 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2172 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2174 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2175 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2178 /* Returns true if 'rule' should be hidden from the controller.
2180 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2181 * (e.g. by in-band control) and are intentionally hidden from the
2184 ofproto_rule_is_hidden(const struct rule *rule)
2186 return rule->cr.priority > UINT16_MAX;
2189 static enum oftable_flags
2190 rule_get_flags(const struct rule *rule)
2192 return rule->ofproto->tables[rule->table_id].flags;
2196 rule_is_modifiable(const struct rule *rule)
2198 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2202 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2204 ofconn_send_reply(ofconn, make_echo_reply(oh));
2209 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2211 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2212 struct ofputil_switch_features features;
2213 struct ofport *port;
2219 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2221 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2223 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2224 * if present, are always at the end.) */
2225 n_tables = ofproto->n_tables;
2226 for (i = 0; i < ofproto->n_tables; i++) {
2227 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2233 features.datapath_id = ofproto->datapath_id;
2234 features.n_buffers = pktbuf_capacity();
2235 features.n_tables = n_tables;
2236 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2237 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2239 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2241 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2242 features.auxiliary_id = 0;
2243 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2245 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2246 ofputil_put_switch_features_port(&port->pp, b);
2249 ofconn_send_reply(ofconn, b);
2254 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2256 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2257 struct ofp_switch_config *osc;
2258 enum ofp_config_flags flags;
2262 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2263 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2264 flags = ofproto->frag_handling;
2265 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2266 if (oh->version < OFP13_VERSION
2267 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2268 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2270 osc->flags = htons(flags);
2271 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2272 ofconn_send_reply(ofconn, buf);
2278 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2280 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2281 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2282 uint16_t flags = ntohs(osc->flags);
2284 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2285 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2286 enum ofp_config_flags cur = ofproto->frag_handling;
2287 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2289 assert((cur & OFPC_FRAG_MASK) == cur);
2291 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2292 ofproto->frag_handling = next;
2294 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2296 ofputil_frag_handling_to_string(next));
2300 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2301 ofconn_set_invalid_ttl_to_controller(ofconn,
2302 (oh->version < OFP13_VERSION
2303 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2305 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2310 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2311 * error message code for the caller to propagate upward. Otherwise, returns
2314 * The log message mentions 'msg_type'. */
2316 reject_slave_controller(struct ofconn *ofconn)
2318 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2319 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2320 return OFPERR_OFPBRC_EPERM;
2327 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2329 struct ofproto *p = ofconn_get_ofproto(ofconn);
2330 struct ofputil_packet_out po;
2331 struct ofpbuf *payload;
2332 uint64_t ofpacts_stub[1024 / 8];
2333 struct ofpbuf ofpacts;
2337 COVERAGE_INC(ofproto_packet_out);
2339 error = reject_slave_controller(ofconn);
2344 /* Decode message. */
2345 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2346 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2348 goto exit_free_ofpacts;
2350 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2351 error = OFPERR_OFPBRC_BAD_PORT;
2352 goto exit_free_ofpacts;
2357 if (po.buffer_id != UINT32_MAX) {
2358 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2359 if (error || !payload) {
2360 goto exit_free_ofpacts;
2363 payload = xmalloc(sizeof *payload);
2364 ofpbuf_use_const(payload, po.packet, po.packet_len);
2367 /* Verify actions against packet, then send packet if successful. */
2368 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2369 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2371 error = p->ofproto_class->packet_out(p, payload, &flow,
2372 po.ofpacts, po.ofpacts_len);
2374 ofpbuf_delete(payload);
2377 ofpbuf_uninit(&ofpacts);
2383 update_port_config(struct ofport *port,
2384 enum ofputil_port_config config,
2385 enum ofputil_port_config mask)
2387 enum ofputil_port_config old_config = port->pp.config;
2388 enum ofputil_port_config toggle;
2390 toggle = (config ^ port->pp.config) & mask;
2391 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2392 if (config & OFPUTIL_PC_PORT_DOWN) {
2393 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2395 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2397 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2400 port->pp.config ^= toggle;
2401 if (port->pp.config != old_config) {
2402 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2407 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2409 struct ofproto *p = ofconn_get_ofproto(ofconn);
2410 struct ofputil_port_mod pm;
2411 struct ofport *port;
2414 error = reject_slave_controller(ofconn);
2419 error = ofputil_decode_port_mod(oh, &pm);
2424 port = ofproto_get_port(p, pm.port_no);
2426 return OFPERR_OFPPMFC_BAD_PORT;
2427 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2428 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2430 update_port_config(port, pm.config, pm.mask);
2432 netdev_set_advertisements(port->netdev, pm.advertise);
2439 handle_desc_stats_request(struct ofconn *ofconn,
2440 const struct ofp_header *request)
2442 struct ofproto *p = ofconn_get_ofproto(ofconn);
2443 struct ofp_desc_stats *ods;
2446 msg = ofpraw_alloc_stats_reply(request, 0);
2447 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2448 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2449 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2450 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2451 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2452 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2453 ofconn_send_reply(ofconn, msg);
2459 handle_table_stats_request(struct ofconn *ofconn,
2460 const struct ofp_header *request)
2462 struct ofproto *p = ofconn_get_ofproto(ofconn);
2463 struct ofp12_table_stats *ots;
2468 /* Set up default values.
2470 * ofp12_table_stats is used as a generic structure as
2471 * it is able to hold all the fields for ofp10_table_stats
2472 * and ofp11_table_stats (and of course itself).
2474 ots = xcalloc(p->n_tables, sizeof *ots);
2475 for (i = 0; i < p->n_tables; i++) {
2476 ots[i].table_id = i;
2477 sprintf(ots[i].name, "table%zu", i);
2478 ots[i].match = htonll(OFPXMT12_MASK);
2479 ots[i].wildcards = htonll(OFPXMT12_MASK);
2480 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2481 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2482 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2483 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2484 ots[i].metadata_match = htonll(UINT64_MAX);
2485 ots[i].metadata_write = htonll(UINT64_MAX);
2486 ots[i].instructions = htonl(OFPIT11_ALL);
2487 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2488 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2489 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2492 p->ofproto_class->get_tables(p, ots);
2494 /* Post-process the tables, dropping hidden tables. */
2495 n_tables = p->n_tables;
2496 for (i = 0; i < p->n_tables; i++) {
2497 const struct oftable *table = &p->tables[i];
2499 if (table->flags & OFTABLE_HIDDEN) {
2505 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2508 if (table->max_flows < ntohl(ots[i].max_entries)) {
2509 ots[i].max_entries = htonl(table->max_flows);
2513 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2514 ofconn_send_reply(ofconn, msg);
2522 append_port_stat(struct ofport *port, struct list *replies)
2524 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2526 /* Intentionally ignore return value, since errors will set
2527 * 'stats' to all-1s, which is correct for OpenFlow, and
2528 * netdev_get_stats() will log errors. */
2529 ofproto_port_get_stats(port, &ops.stats);
2531 ofputil_append_port_stat(replies, &ops);
2535 handle_port_stats_request(struct ofconn *ofconn,
2536 const struct ofp_header *request)
2538 struct ofproto *p = ofconn_get_ofproto(ofconn);
2539 struct ofport *port;
2540 struct list replies;
2544 error = ofputil_decode_port_stats_request(request, &port_no);
2549 ofpmp_init(&replies, request);
2550 if (port_no != OFPP_ANY) {
2551 port = ofproto_get_port(p, port_no);
2553 append_port_stat(port, &replies);
2556 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2557 append_port_stat(port, &replies);
2561 ofconn_send_replies(ofconn, &replies);
2566 handle_port_desc_stats_request(struct ofconn *ofconn,
2567 const struct ofp_header *request)
2569 struct ofproto *p = ofconn_get_ofproto(ofconn);
2570 enum ofp_version version;
2571 struct ofport *port;
2572 struct list replies;
2574 ofpmp_init(&replies, request);
2576 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2577 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2578 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2581 ofconn_send_replies(ofconn, &replies);
2586 calc_flow_duration__(long long int start, long long int now,
2587 uint32_t *sec, uint32_t *nsec)
2589 long long int msecs = now - start;
2590 *sec = msecs / 1000;
2591 *nsec = (msecs % 1000) * (1000 * 1000);
2594 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2595 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2597 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2599 return (table_id == 0xff || table_id < ofproto->n_tables
2601 : OFPERR_OFPBRC_BAD_TABLE_ID);
2605 static struct oftable *
2606 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2608 struct oftable *table;
2610 for (table = &ofproto->tables[table_id];
2611 table < &ofproto->tables[ofproto->n_tables];
2613 if (!(table->flags & OFTABLE_HIDDEN)) {
2621 static struct oftable *
2622 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2624 if (table_id == 0xff) {
2625 return next_visible_table(ofproto, 0);
2626 } else if (table_id < ofproto->n_tables) {
2627 return &ofproto->tables[table_id];
2633 static struct oftable *
2634 next_matching_table(const struct ofproto *ofproto,
2635 const struct oftable *table, uint8_t table_id)
2637 return (table_id == 0xff
2638 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2642 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2644 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2645 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2647 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2648 * only once, for that table. (This can be used to access tables marked
2651 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2652 * entered at all. (Perhaps you should have validated TABLE_ID with
2653 * check_table_id().)
2655 * All parameters are evaluated multiple times.
2657 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2658 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2660 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2662 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2663 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2664 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2667 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2668 * to 'out_port' are included.
2670 * Hidden rules are always omitted.
2672 * Returns 0 on success, otherwise an OpenFlow error code. */
2674 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2675 const struct match *match,
2676 ovs_be64 cookie, ovs_be64 cookie_mask,
2677 uint16_t out_port, struct list *rules)
2679 struct oftable *table;
2683 error = check_table_id(ofproto, table_id);
2689 cls_rule_init(&cr, match, 0);
2690 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2691 struct cls_cursor cursor;
2694 cls_cursor_init(&cursor, &table->cls, &cr);
2695 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2696 if (rule->pending) {
2697 error = OFPROTO_POSTPONE;
2700 if (!ofproto_rule_is_hidden(rule)
2701 && ofproto_rule_has_out_port(rule, out_port)
2702 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2703 list_push_back(rules, &rule->ofproto_node);
2709 cls_rule_destroy(&cr);
2713 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2714 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2715 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2718 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2719 * to 'out_port' are included.
2721 * Hidden rules are always omitted.
2723 * Returns 0 on success, otherwise an OpenFlow error code. */
2725 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2726 const struct match *match, unsigned int priority,
2727 ovs_be64 cookie, ovs_be64 cookie_mask,
2728 uint16_t out_port, struct list *rules)
2730 struct oftable *table;
2734 error = check_table_id(ofproto, table_id);
2740 cls_rule_init(&cr, match, priority);
2741 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2744 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2747 if (rule->pending) {
2748 error = OFPROTO_POSTPONE;
2751 if (!ofproto_rule_is_hidden(rule)
2752 && ofproto_rule_has_out_port(rule, out_port)
2753 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2754 list_push_back(rules, &rule->ofproto_node);
2760 cls_rule_destroy(&cr);
2764 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2765 * forced into the range of a uint16_t. */
2767 age_secs(long long int age_ms)
2769 return (age_ms < 0 ? 0
2770 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2771 : (unsigned int) age_ms / 1000);
2775 handle_flow_stats_request(struct ofconn *ofconn,
2776 const struct ofp_header *request)
2778 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2779 struct ofputil_flow_stats_request fsr;
2780 struct list replies;
2785 error = ofputil_decode_flow_stats_request(&fsr, request);
2790 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2791 fsr.cookie, fsr.cookie_mask,
2792 fsr.out_port, &rules);
2797 ofpmp_init(&replies, request);
2798 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2799 long long int now = time_msec();
2800 struct ofputil_flow_stats fs;
2802 minimatch_expand(&rule->cr.match, &fs.match);
2803 fs.priority = rule->cr.priority;
2804 fs.cookie = rule->flow_cookie;
2805 fs.table_id = rule->table_id;
2806 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2808 fs.idle_timeout = rule->idle_timeout;
2809 fs.hard_timeout = rule->hard_timeout;
2810 fs.idle_age = age_secs(now - rule->used);
2811 fs.hard_age = age_secs(now - rule->modified);
2812 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2814 fs.ofpacts = rule->ofpacts;
2815 fs.ofpacts_len = rule->ofpacts_len;
2817 if (rule->send_flow_removed) {
2818 fs.flags |= OFPFF_SEND_FLOW_REM;
2819 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2820 and OFPFF13_NO_BYT_COUNTS */
2822 ofputil_append_flow_stats_reply(&fs, &replies);
2824 ofconn_send_replies(ofconn, &replies);
2830 flow_stats_ds(struct rule *rule, struct ds *results)
2832 uint64_t packet_count, byte_count;
2834 rule->ofproto->ofproto_class->rule_get_stats(rule,
2835 &packet_count, &byte_count);
2837 if (rule->table_id != 0) {
2838 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2840 ds_put_format(results, "duration=%llds, ",
2841 (time_msec() - rule->created) / 1000);
2842 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2843 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2844 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2845 cls_rule_format(&rule->cr, results);
2846 ds_put_char(results, ',');
2847 if (rule->ofpacts_len > 0) {
2848 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2850 ds_put_cstr(results, "drop");
2852 ds_put_cstr(results, "\n");
2855 /* Adds a pretty-printed description of all flows to 'results', including
2856 * hidden flows (e.g., set up by in-band control). */
2858 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2860 struct oftable *table;
2862 OFPROTO_FOR_EACH_TABLE (table, p) {
2863 struct cls_cursor cursor;
2866 cls_cursor_init(&cursor, &table->cls, NULL);
2867 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2868 flow_stats_ds(rule, results);
2873 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2874 * '*engine_type' and '*engine_id', respectively. */
2876 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2877 uint8_t *engine_type, uint8_t *engine_id)
2879 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2882 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2883 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2884 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2885 * and -1 if CFM is not enabled on 'ofp_port'. */
2887 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2889 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2890 return (ofport && ofproto->ofproto_class->get_cfm_fault
2891 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2895 /* Checks the operational status reported by the remote CFM endpoint of
2896 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2897 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2900 ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
2902 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2903 return (ofport && ofproto->ofproto_class->get_cfm_opup
2904 ? ofproto->ofproto_class->get_cfm_opup(ofport)
2908 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2909 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2910 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2911 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2913 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2914 uint16_t ofp_port, const uint64_t **rmps,
2917 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2921 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2922 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2927 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2928 * integer value between 0 and 100 to indicate the health of the port as a
2929 * percentage which is the average of cfm health of all the remote_mpids or
2930 * returns -1 if CFM is not enabled on 'ofport'. */
2932 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2934 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2935 return (ofport && ofproto->ofproto_class->get_cfm_health
2936 ? ofproto->ofproto_class->get_cfm_health(ofport)
2941 handle_aggregate_stats_request(struct ofconn *ofconn,
2942 const struct ofp_header *oh)
2944 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2945 struct ofputil_flow_stats_request request;
2946 struct ofputil_aggregate_stats stats;
2947 bool unknown_packets, unknown_bytes;
2948 struct ofpbuf *reply;
2953 error = ofputil_decode_flow_stats_request(&request, oh);
2958 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2959 request.cookie, request.cookie_mask,
2960 request.out_port, &rules);
2965 memset(&stats, 0, sizeof stats);
2966 unknown_packets = unknown_bytes = false;
2967 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2968 uint64_t packet_count;
2969 uint64_t byte_count;
2971 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2974 if (packet_count == UINT64_MAX) {
2975 unknown_packets = true;
2977 stats.packet_count += packet_count;
2980 if (byte_count == UINT64_MAX) {
2981 unknown_bytes = true;
2983 stats.byte_count += byte_count;
2988 if (unknown_packets) {
2989 stats.packet_count = UINT64_MAX;
2991 if (unknown_bytes) {
2992 stats.byte_count = UINT64_MAX;
2995 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2996 ofconn_send_reply(ofconn, reply);
3001 struct queue_stats_cbdata {
3002 struct ofport *ofport;
3003 struct list replies;
3007 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3008 const struct netdev_queue_stats *stats)
3011 struct ofputil_queue_stats oqs = {
3012 .port_no = cbdata->ofport->pp.port_no,
3013 .queue_id = queue_id,
3016 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3020 handle_queue_stats_dump_cb(uint32_t queue_id,
3021 struct netdev_queue_stats *stats,
3024 struct queue_stats_cbdata *cbdata = cbdata_;
3026 put_queue_stats(cbdata, queue_id, stats);
3030 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3031 struct queue_stats_cbdata *cbdata)
3033 cbdata->ofport = port;
3034 if (queue_id == OFPQ_ALL) {
3035 netdev_dump_queue_stats(port->netdev,
3036 handle_queue_stats_dump_cb, cbdata);
3038 struct netdev_queue_stats stats;
3040 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3041 put_queue_stats(cbdata, queue_id, &stats);
3043 return OFPERR_OFPQOFC_BAD_QUEUE;
3050 handle_queue_stats_request(struct ofconn *ofconn,
3051 const struct ofp_header *rq)
3053 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3054 struct queue_stats_cbdata cbdata;
3055 struct ofport *port;
3057 struct ofputil_queue_stats_request oqsr;
3059 COVERAGE_INC(ofproto_queue_req);
3061 ofpmp_init(&cbdata.replies, rq);
3063 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3068 if (oqsr.port_no == OFPP_ANY) {
3069 error = OFPERR_OFPQOFC_BAD_QUEUE;
3070 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3071 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3076 port = ofproto_get_port(ofproto, oqsr.port_no);
3078 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3079 : OFPERR_OFPQOFC_BAD_PORT);
3082 ofconn_send_replies(ofconn, &cbdata.replies);
3084 ofpbuf_list_delete(&cbdata.replies);
3091 is_flow_deletion_pending(const struct ofproto *ofproto,
3092 const struct cls_rule *cls_rule,
3095 if (!hmap_is_empty(&ofproto->deletions)) {
3096 struct ofoperation *op;
3098 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3099 cls_rule_hash(cls_rule, table_id),
3100 &ofproto->deletions) {
3101 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3110 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3111 * in which no matching flow already exists in the flow table.
3113 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3114 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3115 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3116 * initiated now but may be retried later.
3118 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3119 * ownership remains with the caller.
3121 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3124 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3125 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3127 struct oftable *table;
3128 struct ofopgroup *group;
3129 struct rule *victim;
3134 error = check_table_id(ofproto, fm->table_id);
3140 if (fm->table_id == 0xff) {
3142 if (ofproto->ofproto_class->rule_choose_table) {
3143 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3149 assert(table_id < ofproto->n_tables);
3150 table = &ofproto->tables[table_id];
3152 table = &ofproto->tables[0];
3154 } else if (fm->table_id < ofproto->n_tables) {
3155 table = &ofproto->tables[fm->table_id];
3157 return OFPERR_OFPBRC_BAD_TABLE_ID;
3160 if (table->flags & OFTABLE_READONLY) {
3161 return OFPERR_OFPBRC_EPERM;
3164 /* Allocate new rule and initialize classifier rule. */
3165 rule = ofproto->ofproto_class->rule_alloc();
3167 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3168 ofproto->name, strerror(error));
3171 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3173 /* Serialize against pending deletion. */
3174 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3175 cls_rule_destroy(&rule->cr);
3176 ofproto->ofproto_class->rule_dealloc(rule);
3177 return OFPROTO_POSTPONE;
3180 /* Check for overlap, if requested. */
3181 if (fm->flags & OFPFF_CHECK_OVERLAP
3182 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3183 cls_rule_destroy(&rule->cr);
3184 ofproto->ofproto_class->rule_dealloc(rule);
3185 return OFPERR_OFPFMFC_OVERLAP;
3188 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3190 rule->ofproto = ofproto;
3191 rule->pending = NULL;
3192 rule->flow_cookie = fm->new_cookie;
3193 rule->created = rule->modified = rule->used = time_msec();
3194 rule->idle_timeout = fm->idle_timeout;
3195 rule->hard_timeout = fm->hard_timeout;
3196 rule->table_id = table - ofproto->tables;
3197 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3198 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3199 and OFPFF13_NO_BYT_COUNTS */
3200 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3201 rule->ofpacts_len = fm->ofpacts_len;
3202 rule->evictable = true;
3203 rule->eviction_group = NULL;
3204 rule->monitor_flags = 0;
3205 rule->add_seqno = 0;
3206 rule->modify_seqno = 0;
3208 /* Insert new rule. */
3209 victim = oftable_replace_rule(rule);
3210 if (victim && !rule_is_modifiable(victim)) {
3211 error = OFPERR_OFPBRC_EPERM;
3212 } else if (victim && victim->pending) {
3213 error = OFPROTO_POSTPONE;
3215 struct ofoperation *op;
3218 if (classifier_count(&table->cls) > table->max_flows) {
3221 was_evictable = rule->evictable;
3222 rule->evictable = false;
3223 evict = choose_rule_to_evict(table);
3224 rule->evictable = was_evictable;
3227 error = OFPERR_OFPFMFC_TABLE_FULL;
3229 } else if (evict->pending) {
3230 error = OFPROTO_POSTPONE;
3237 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3238 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3239 op->victim = victim;
3241 error = ofproto->ofproto_class->rule_construct(rule);
3243 op->group->n_running--;
3244 ofoperation_destroy(rule->pending);
3246 delete_flow__(evict, group);
3248 ofopgroup_submit(group);
3252 /* Back out if an error occurred. */
3254 oftable_substitute_rule(rule, victim);
3255 ofproto_rule_destroy__(rule);
3260 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3262 /* Modifies the rules listed in 'rules', changing their actions to match those
3265 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3268 * Returns 0 on success, otherwise an OpenFlow error code. */
3270 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3271 const struct ofputil_flow_mod *fm,
3272 const struct ofp_header *request, struct list *rules)
3274 struct ofopgroup *group;
3278 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3279 error = OFPERR_OFPBRC_EPERM;
3280 LIST_FOR_EACH (rule, ofproto_node, rules) {
3281 struct ofoperation *op;
3282 bool actions_changed;
3283 ovs_be64 new_cookie;
3285 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3287 if (rule_is_modifiable(rule)) {
3288 /* At least one rule is modifiable, don't report EPERM error. */
3294 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3295 rule->ofpacts, rule->ofpacts_len);
3296 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3298 : rule->flow_cookie);
3299 if (!actions_changed && new_cookie == rule->flow_cookie) {
3300 /* No change at all. */
3304 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3305 rule->flow_cookie = new_cookie;
3306 if (actions_changed) {
3307 op->ofpacts = rule->ofpacts;
3308 op->ofpacts_len = rule->ofpacts_len;
3309 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3310 rule->ofpacts_len = fm->ofpacts_len;
3311 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3313 ofoperation_complete(op, 0);
3316 ofopgroup_submit(group);
3322 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3323 const struct ofputil_flow_mod *fm,
3324 const struct ofp_header *request)
3326 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3329 return add_flow(ofproto, ofconn, fm, request);
3332 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3335 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3338 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3339 const struct ofputil_flow_mod *fm,
3340 const struct ofp_header *request)
3345 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3346 fm->cookie, fm->cookie_mask,
3350 } else if (list_is_empty(&rules)) {
3351 return modify_flows_add(ofproto, ofconn, fm, request);
3353 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3357 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3360 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3363 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3364 const struct ofputil_flow_mod *fm,
3365 const struct ofp_header *request)
3370 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3371 fm->priority, fm->cookie, fm->cookie_mask,
3376 } else if (list_is_empty(&rules)) {
3377 return modify_flows_add(ofproto, ofconn, fm, request);
3379 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3380 fm, request, &rules)
3385 /* OFPFC_DELETE implementation. */
3388 delete_flow__(struct rule *rule, struct ofopgroup *group)
3390 struct ofproto *ofproto = rule->ofproto;
3392 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3394 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3395 oftable_remove_rule(rule);
3396 ofproto->ofproto_class->rule_destruct(rule);
3399 /* Deletes the rules listed in 'rules'.
3401 * Returns 0 on success, otherwise an OpenFlow error code. */
3403 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3404 const struct ofp_header *request, struct list *rules)
3406 struct rule *rule, *next;
3407 struct ofopgroup *group;
3409 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3410 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3411 delete_flow__(rule, group);
3413 ofopgroup_submit(group);
3418 /* Implements OFPFC_DELETE. */
3420 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3421 const struct ofputil_flow_mod *fm,
3422 const struct ofp_header *request)
3427 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3428 fm->cookie, fm->cookie_mask,
3429 fm->out_port, &rules);
3430 return (error ? error
3431 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3436 /* Implements OFPFC_DELETE_STRICT. */
3438 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3439 const struct ofputil_flow_mod *fm,
3440 const struct ofp_header *request)
3445 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3446 fm->priority, fm->cookie, fm->cookie_mask,
3447 fm->out_port, &rules);
3448 return (error ? error
3449 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3455 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3457 struct ofputil_flow_removed fr;
3459 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3463 minimatch_expand(&rule->cr.match, &fr.match);
3464 fr.priority = rule->cr.priority;
3465 fr.cookie = rule->flow_cookie;
3467 fr.table_id = rule->table_id;
3468 calc_flow_duration__(rule->created, time_msec(),
3469 &fr.duration_sec, &fr.duration_nsec);
3470 fr.idle_timeout = rule->idle_timeout;
3471 fr.hard_timeout = rule->hard_timeout;
3472 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3475 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3479 ofproto_rule_update_used(struct rule *rule, long long int used)
3481 if (used > rule->used) {
3482 struct eviction_group *evg = rule->eviction_group;
3486 heap_change(&evg->rules, &rule->evg_node,
3487 rule_eviction_priority(rule));
3492 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3493 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3496 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3499 * ofproto implementation ->run() functions should use this function to expire
3500 * OpenFlow flows. */
3502 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3504 struct ofproto *ofproto = rule->ofproto;
3505 struct ofopgroup *group;
3507 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3509 ofproto_rule_send_removed(rule, reason);
3511 group = ofopgroup_create_unattached(ofproto);
3512 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3513 oftable_remove_rule(rule);
3514 ofproto->ofproto_class->rule_destruct(rule);
3515 ofopgroup_submit(group);
3519 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3521 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3522 struct ofputil_flow_mod fm;
3523 uint64_t ofpacts_stub[1024 / 8];
3524 struct ofpbuf ofpacts;
3528 error = reject_slave_controller(ofconn);
3533 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3534 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3537 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3538 &fm.match.flow, ofproto->max_ports);
3541 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3544 goto exit_free_ofpacts;
3547 /* Record the operation for logging a summary report. */
3548 switch (fm.command) {
3554 case OFPFC_MODIFY_STRICT:
3555 ofproto->n_modify++;
3559 case OFPFC_DELETE_STRICT:
3560 ofproto->n_delete++;
3565 if (ofproto->next_op_report == LLONG_MAX) {
3566 ofproto->first_op = now;
3567 ofproto->next_op_report = MAX(now + 10 * 1000,
3568 ofproto->op_backoff);
3569 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3571 ofproto->last_op = now;
3574 ofpbuf_uninit(&ofpacts);
3580 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3581 const struct ofputil_flow_mod *fm,
3582 const struct ofp_header *oh)
3584 if (ofproto->n_pending >= 50) {
3585 assert(!list_is_empty(&ofproto->pending));
3586 return OFPROTO_POSTPONE;
3589 switch (fm->command) {
3591 return add_flow(ofproto, ofconn, fm, oh);
3594 return modify_flows_loose(ofproto, ofconn, fm, oh);
3596 case OFPFC_MODIFY_STRICT:
3597 return modify_flow_strict(ofproto, ofconn, fm, oh);
3600 return delete_flows_loose(ofproto, ofconn, fm, oh);
3602 case OFPFC_DELETE_STRICT:
3603 return delete_flow_strict(ofproto, ofconn, fm, oh);
3606 if (fm->command > 0xff) {
3607 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3608 "flow_mod_table_id extension is not enabled",
3611 return OFPERR_OFPFMFC_BAD_COMMAND;
3616 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3618 const struct nx_role_request *nrr = ofpmsg_body(oh);
3619 struct nx_role_request *reply;
3623 role = ntohl(nrr->role);
3624 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3625 && role != NX_ROLE_SLAVE) {
3626 return OFPERR_OFPRRFC_BAD_ROLE;
3629 if (ofconn_get_role(ofconn) != role
3630 && ofconn_has_pending_opgroups(ofconn)) {
3631 return OFPROTO_POSTPONE;
3634 ofconn_set_role(ofconn, role);
3636 buf = ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY, oh, 0);
3637 reply = ofpbuf_put_zeros(buf, sizeof *reply);
3638 reply->role = htonl(role);
3639 ofconn_send_reply(ofconn, buf);
3645 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3646 const struct ofp_header *oh)
3648 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3649 enum ofputil_protocol cur, next;
3651 cur = ofconn_get_protocol(ofconn);
3652 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3653 ofconn_set_protocol(ofconn, next);
3659 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3661 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3662 enum ofputil_protocol cur, next;
3663 enum ofputil_protocol next_base;
3665 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3667 return OFPERR_OFPBRC_EPERM;
3670 cur = ofconn_get_protocol(ofconn);
3671 next = ofputil_protocol_set_base(cur, next_base);
3672 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3673 /* Avoid sending async messages in surprising protocol. */
3674 return OFPROTO_POSTPONE;
3677 ofconn_set_protocol(ofconn, next);
3682 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3683 const struct ofp_header *oh)
3685 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3688 format = ntohl(msg->format);
3689 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3690 return OFPERR_OFPBRC_EPERM;
3693 if (format != ofconn_get_packet_in_format(ofconn)
3694 && ofconn_has_pending_opgroups(ofconn)) {
3695 /* Avoid sending async message in surprsing packet in format. */
3696 return OFPROTO_POSTPONE;
3699 ofconn_set_packet_in_format(ofconn, format);
3704 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3706 const struct nx_async_config *msg = ofpmsg_body(oh);
3707 uint32_t master[OAM_N_TYPES];
3708 uint32_t slave[OAM_N_TYPES];
3710 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3711 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3712 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3714 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3715 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3716 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3718 ofconn_set_async_config(ofconn, master, slave);
3719 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3720 !ofconn_get_miss_send_len(ofconn)) {
3721 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3728 handle_nxt_set_controller_id(struct ofconn *ofconn,
3729 const struct ofp_header *oh)
3731 const struct nx_controller_id *nci = ofpmsg_body(oh);
3733 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3734 return OFPERR_NXBRC_MUST_BE_ZERO;
3737 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3742 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3746 if (ofconn_has_pending_opgroups(ofconn)) {
3747 return OFPROTO_POSTPONE;
3750 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3751 ? OFPRAW_OFPT10_BARRIER_REPLY
3752 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3753 ofconn_send_reply(ofconn, buf);
3758 ofproto_compose_flow_refresh_update(const struct rule *rule,
3759 enum nx_flow_monitor_flags flags,
3762 struct ofoperation *op = rule->pending;
3763 struct ofputil_flow_update fu;
3766 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3767 /* We'll report the final flow when the operation completes. Reporting
3768 * it now would cause a duplicate report later. */
3772 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3773 ? NXFME_ADDED : NXFME_MODIFIED);
3775 fu.idle_timeout = rule->idle_timeout;
3776 fu.hard_timeout = rule->hard_timeout;
3777 fu.table_id = rule->table_id;
3778 fu.cookie = rule->flow_cookie;
3779 minimatch_expand(&rule->cr.match, &match);
3781 fu.priority = rule->cr.priority;
3782 if (!(flags & NXFMF_ACTIONS)) {
3786 fu.ofpacts = rule->ofpacts;
3787 fu.ofpacts_len = rule->ofpacts_len;
3789 /* An operation is in progress. Use the previous version of the flow's
3790 * actions, so that when the operation commits we report the change. */
3792 case OFOPERATION_ADD:
3793 /* We already verified that there was a victim. */
3794 fu.ofpacts = op->victim->ofpacts;
3795 fu.ofpacts_len = op->victim->ofpacts_len;
3798 case OFOPERATION_MODIFY:
3800 fu.ofpacts = op->ofpacts;
3801 fu.ofpacts_len = op->ofpacts_len;
3803 fu.ofpacts = rule->ofpacts;
3804 fu.ofpacts_len = rule->ofpacts_len;
3808 case OFOPERATION_DELETE:
3809 fu.ofpacts = rule->ofpacts;
3810 fu.ofpacts_len = rule->ofpacts_len;
3818 if (list_is_empty(msgs)) {
3819 ofputil_start_flow_update(msgs);
3821 ofputil_append_flow_update(&fu, msgs);
3825 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3829 LIST_FOR_EACH (rule, ofproto_node, rules) {
3830 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3831 rule->monitor_flags = 0;
3833 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3838 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3839 struct rule *rule, uint64_t seqno,
3842 enum nx_flow_monitor_flags update;
3844 if (ofproto_rule_is_hidden(rule)) {
3849 ? ofoperation_has_out_port(rule->pending, m->out_port)
3850 : ofproto_rule_has_out_port(rule, m->out_port))) {
3855 if (rule->add_seqno > seqno) {
3856 update = NXFMF_ADD | NXFMF_MODIFY;
3857 } else if (rule->modify_seqno > seqno) {
3858 update = NXFMF_MODIFY;
3863 if (!(m->flags & update)) {
3867 update = NXFMF_INITIAL;
3870 if (!rule->monitor_flags) {
3871 list_push_back(rules, &rule->ofproto_node);
3873 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3877 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3881 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3882 const struct ofoperation *op;
3883 const struct oftable *table;
3884 struct cls_rule target;
3886 cls_rule_init_from_minimatch(&target, &m->match, 0);
3887 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3888 struct cls_cursor cursor;
3891 cls_cursor_init(&cursor, &table->cls, &target);
3892 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3893 assert(!rule->pending); /* XXX */
3894 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3898 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3899 struct rule *rule = op->rule;
3901 if (((m->table_id == 0xff
3902 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3903 : m->table_id == rule->table_id))
3904 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3905 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3908 cls_rule_destroy(&target);
3912 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3915 if (m->flags & NXFMF_INITIAL) {
3916 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3921 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3922 uint64_t seqno, struct list *rules)
3924 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3928 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3930 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3931 struct ofmonitor **monitors;
3932 size_t n_monitors, allocated_monitors;
3933 struct list replies;
3940 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3942 n_monitors = allocated_monitors = 0;
3944 struct ofputil_flow_monitor_request request;
3945 struct ofmonitor *m;
3948 retval = ofputil_decode_flow_monitor_request(&request, &b);
3949 if (retval == EOF) {
3951 } else if (retval) {
3956 if (request.table_id != 0xff
3957 && request.table_id >= ofproto->n_tables) {
3958 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3962 error = ofmonitor_create(&request, ofconn, &m);
3967 if (n_monitors >= allocated_monitors) {
3968 monitors = x2nrealloc(monitors, &allocated_monitors,
3971 monitors[n_monitors++] = m;
3975 for (i = 0; i < n_monitors; i++) {
3976 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3979 ofpmp_init(&replies, oh);
3980 ofmonitor_compose_refresh_updates(&rules, &replies);
3981 ofconn_send_replies(ofconn, &replies);
3988 for (i = 0; i < n_monitors; i++) {
3989 ofmonitor_destroy(monitors[i]);
3996 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3998 struct ofmonitor *m;
4001 id = ofputil_decode_flow_monitor_cancel(oh);
4002 m = ofmonitor_lookup(ofconn, id);
4004 return OFPERR_NXBRC_FM_BAD_ID;
4007 ofmonitor_destroy(m);
4012 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4014 const struct ofp_header *oh = msg->data;
4018 error = ofptype_decode(&type, oh);
4024 /* OpenFlow requests. */
4025 case OFPTYPE_ECHO_REQUEST:
4026 return handle_echo_request(ofconn, oh);
4028 case OFPTYPE_FEATURES_REQUEST:
4029 return handle_features_request(ofconn, oh);
4031 case OFPTYPE_GET_CONFIG_REQUEST:
4032 return handle_get_config_request(ofconn, oh);
4034 case OFPTYPE_SET_CONFIG:
4035 return handle_set_config(ofconn, oh);
4037 case OFPTYPE_PACKET_OUT:
4038 return handle_packet_out(ofconn, oh);
4040 case OFPTYPE_PORT_MOD:
4041 return handle_port_mod(ofconn, oh);
4043 case OFPTYPE_FLOW_MOD:
4044 return handle_flow_mod(ofconn, oh);
4046 case OFPTYPE_BARRIER_REQUEST:
4047 return handle_barrier_request(ofconn, oh);
4049 /* OpenFlow replies. */
4050 case OFPTYPE_ECHO_REPLY:
4053 /* Nicira extension requests. */
4054 case OFPTYPE_ROLE_REQUEST:
4055 return handle_role_request(ofconn, oh);
4057 case OFPTYPE_FLOW_MOD_TABLE_ID:
4058 return handle_nxt_flow_mod_table_id(ofconn, oh);
4060 case OFPTYPE_SET_FLOW_FORMAT:
4061 return handle_nxt_set_flow_format(ofconn, oh);
4063 case OFPTYPE_SET_PACKET_IN_FORMAT:
4064 return handle_nxt_set_packet_in_format(ofconn, oh);
4066 case OFPTYPE_SET_CONTROLLER_ID:
4067 return handle_nxt_set_controller_id(ofconn, oh);
4069 case OFPTYPE_FLOW_AGE:
4070 /* Nothing to do. */
4073 case OFPTYPE_FLOW_MONITOR_CANCEL:
4074 return handle_flow_monitor_cancel(ofconn, oh);
4076 case OFPTYPE_SET_ASYNC_CONFIG:
4077 return handle_nxt_set_async_config(ofconn, oh);
4079 /* Statistics requests. */
4080 case OFPTYPE_DESC_STATS_REQUEST:
4081 return handle_desc_stats_request(ofconn, oh);
4083 case OFPTYPE_FLOW_STATS_REQUEST:
4084 return handle_flow_stats_request(ofconn, oh);
4086 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4087 return handle_aggregate_stats_request(ofconn, oh);
4089 case OFPTYPE_TABLE_STATS_REQUEST:
4090 return handle_table_stats_request(ofconn, oh);
4092 case OFPTYPE_PORT_STATS_REQUEST:
4093 return handle_port_stats_request(ofconn, oh);
4095 case OFPTYPE_QUEUE_STATS_REQUEST:
4096 return handle_queue_stats_request(ofconn, oh);
4098 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4099 return handle_port_desc_stats_request(ofconn, oh);
4101 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4102 return handle_flow_monitor_request(ofconn, oh);
4104 /* FIXME: Change the following once they are implemented: */
4105 case OFPTYPE_GET_ASYNC_REQUEST:
4106 case OFPTYPE_METER_MOD:
4107 case OFPTYPE_GROUP_REQUEST:
4108 case OFPTYPE_GROUP_DESC_REQUEST:
4109 case OFPTYPE_GROUP_FEATURES_REQUEST:
4110 case OFPTYPE_METER_REQUEST:
4111 case OFPTYPE_METER_CONFIG_REQUEST:
4112 case OFPTYPE_METER_FEATURES_REQUEST:
4113 case OFPTYPE_TABLE_FEATURES_REQUEST:
4114 return OFPERR_OFPBRC_BAD_TYPE;
4118 case OFPTYPE_FEATURES_REPLY:
4119 case OFPTYPE_GET_CONFIG_REPLY:
4120 case OFPTYPE_PACKET_IN:
4121 case OFPTYPE_FLOW_REMOVED:
4122 case OFPTYPE_PORT_STATUS:
4123 case OFPTYPE_BARRIER_REPLY:
4124 case OFPTYPE_DESC_STATS_REPLY:
4125 case OFPTYPE_FLOW_STATS_REPLY:
4126 case OFPTYPE_QUEUE_STATS_REPLY:
4127 case OFPTYPE_PORT_STATS_REPLY:
4128 case OFPTYPE_TABLE_STATS_REPLY:
4129 case OFPTYPE_AGGREGATE_STATS_REPLY:
4130 case OFPTYPE_PORT_DESC_STATS_REPLY:
4131 case OFPTYPE_ROLE_REPLY:
4132 case OFPTYPE_FLOW_MONITOR_PAUSED:
4133 case OFPTYPE_FLOW_MONITOR_RESUMED:
4134 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4135 case OFPTYPE_GET_ASYNC_REPLY:
4136 case OFPTYPE_GROUP_REPLY:
4137 case OFPTYPE_GROUP_DESC_REPLY:
4138 case OFPTYPE_GROUP_FEATURES_REPLY:
4139 case OFPTYPE_METER_REPLY:
4140 case OFPTYPE_METER_CONFIG_REPLY:
4141 case OFPTYPE_METER_FEATURES_REPLY:
4142 case OFPTYPE_TABLE_FEATURES_REPLY:
4144 return OFPERR_OFPBRC_BAD_TYPE;
4149 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4151 int error = handle_openflow__(ofconn, ofp_msg);
4152 if (error && error != OFPROTO_POSTPONE) {
4153 ofconn_send_error(ofconn, ofp_msg->data, error);
4155 COVERAGE_INC(ofproto_recv_openflow);
4156 return error != OFPROTO_POSTPONE;
4159 /* Asynchronous operations. */
4161 /* Creates and returns a new ofopgroup that is not associated with any
4162 * OpenFlow connection.
4164 * The caller should add operations to the returned group with
4165 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4166 static struct ofopgroup *
4167 ofopgroup_create_unattached(struct ofproto *ofproto)
4169 struct ofopgroup *group = xzalloc(sizeof *group);
4170 group->ofproto = ofproto;
4171 list_init(&group->ofproto_node);
4172 list_init(&group->ops);
4173 list_init(&group->ofconn_node);
4177 /* Creates and returns a new ofopgroup for 'ofproto'.
4179 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4180 * connection. The 'request' and 'buffer_id' arguments are ignored.
4182 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4183 * If the ofopgroup eventually fails, then the error reply will include
4184 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4185 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4187 * The caller should add operations to the returned group with
4188 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4189 static struct ofopgroup *
4190 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4191 const struct ofp_header *request, uint32_t buffer_id)
4193 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4195 size_t request_len = ntohs(request->length);
4197 assert(ofconn_get_ofproto(ofconn) == ofproto);
4199 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4200 group->ofconn = ofconn;
4201 group->request = xmemdup(request, MIN(request_len, 64));
4202 group->buffer_id = buffer_id;
4207 /* Submits 'group' for processing.
4209 * If 'group' contains no operations (e.g. none were ever added, or all of the
4210 * ones that were added completed synchronously), then it is destroyed
4211 * immediately. Otherwise it is added to the ofproto's list of pending
4214 ofopgroup_submit(struct ofopgroup *group)
4216 if (!group->n_running) {
4217 ofopgroup_complete(group);
4219 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4220 group->ofproto->n_pending++;
4225 ofopgroup_complete(struct ofopgroup *group)
4227 struct ofproto *ofproto = group->ofproto;
4229 struct ofconn *abbrev_ofconn;
4230 ovs_be32 abbrev_xid;
4232 struct ofoperation *op, *next_op;
4235 assert(!group->n_running);
4238 LIST_FOR_EACH (op, group_node, &group->ops) {
4245 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4246 LIST_FOR_EACH (op, group_node, &group->ops) {
4247 if (op->type != OFOPERATION_DELETE) {
4248 struct ofpbuf *packet;
4251 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4255 error = rule_execute(op->rule, in_port, packet);
4262 if (!error && !list_is_empty(&group->ofconn_node)) {
4263 abbrev_ofconn = group->ofconn;
4264 abbrev_xid = group->request->xid;
4266 abbrev_ofconn = NULL;
4267 abbrev_xid = htonl(0);
4269 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4270 struct rule *rule = op->rule;
4272 if (!op->error && !ofproto_rule_is_hidden(rule)) {
4273 /* Check that we can just cast from ofoperation_type to
4274 * nx_flow_update_event. */
4275 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4277 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4279 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4282 ofmonitor_report(ofproto->connmgr, rule,
4283 (enum nx_flow_update_event) op->type,
4284 op->reason, abbrev_ofconn, abbrev_xid);
4287 rule->pending = NULL;
4290 case OFOPERATION_ADD:
4294 ofproto_rule_destroy__(op->victim);
4295 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4296 if (vid_mask == VLAN_VID_MASK) {
4297 if (ofproto->vlan_bitmap) {
4298 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4299 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4300 bitmap_set1(ofproto->vlan_bitmap, vid);
4301 ofproto->vlans_changed = true;
4304 ofproto->vlans_changed = true;
4308 oftable_substitute_rule(rule, op->victim);
4309 ofproto_rule_destroy__(rule);
4313 case OFOPERATION_DELETE:
4315 ofproto_rule_destroy__(rule);
4319 case OFOPERATION_MODIFY:
4321 rule->modified = time_msec();
4323 rule->flow_cookie = op->flow_cookie;
4325 free(rule->ofpacts);
4326 rule->ofpacts = op->ofpacts;
4327 rule->ofpacts_len = op->ofpacts_len;
4329 op->ofpacts_len = 0;
4338 ofoperation_destroy(op);
4341 ofmonitor_flush(ofproto->connmgr);
4343 if (!list_is_empty(&group->ofproto_node)) {
4344 assert(ofproto->n_pending > 0);
4345 ofproto->n_pending--;
4346 list_remove(&group->ofproto_node);
4348 if (!list_is_empty(&group->ofconn_node)) {
4349 list_remove(&group->ofconn_node);
4351 ofconn_send_error(group->ofconn, group->request, error);
4353 connmgr_retry(ofproto->connmgr);
4355 free(group->request);
4359 /* Initiates a new operation on 'rule', of the specified 'type', within
4360 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4362 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4363 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4365 * Returns the newly created ofoperation (which is also available as
4366 * rule->pending). */
4367 static struct ofoperation *
4368 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4369 enum ofoperation_type type,
4370 enum ofp_flow_removed_reason reason)
4372 struct ofproto *ofproto = group->ofproto;
4373 struct ofoperation *op;
4375 assert(!rule->pending);
4377 op = rule->pending = xzalloc(sizeof *op);
4379 list_push_back(&group->ops, &op->group_node);
4382 op->reason = reason;
4383 op->flow_cookie = rule->flow_cookie;
4387 if (type == OFOPERATION_DELETE) {
4388 hmap_insert(&ofproto->deletions, &op->hmap_node,
4389 cls_rule_hash(&rule->cr, rule->table_id));
4396 ofoperation_destroy(struct ofoperation *op)
4398 struct ofopgroup *group = op->group;
4401 op->rule->pending = NULL;
4403 if (op->type == OFOPERATION_DELETE) {
4404 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4406 list_remove(&op->group_node);
4411 /* Indicates that 'op' completed with status 'error', which is either 0 to
4412 * indicate success or an OpenFlow error code on failure.
4414 * If 'error' is 0, indicating success, the operation will be committed
4415 * permanently to the flow table. There is one interesting subcase:
4417 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4418 * the flow table (the "victim" rule) by a new one, then the caller must
4419 * have uninitialized any derived state in the victim rule, as in step 5 in
4420 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4421 * performs steps 6 and 7 for the victim rule, most notably by calling its
4422 * ->rule_dealloc() function.
4424 * If 'error' is nonzero, then generally the operation will be rolled back:
4426 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4427 * restores the original rule. The caller must have uninitialized any
4428 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4429 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4430 * and 7 for the new rule, calling its ->rule_dealloc() function.
4432 * - If 'op' is a "modify flow" operation, ofproto restores the original
4435 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4436 * allowed to fail. It must always succeed.
4438 * Please see the large comment in ofproto/ofproto-provider.h titled
4439 * "Asynchronous Operation Support" for more information. */
4441 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4443 struct ofopgroup *group = op->group;
4445 assert(op->rule->pending == op);
4446 assert(group->n_running > 0);
4447 assert(!error || op->type != OFOPERATION_DELETE);
4450 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4451 ofopgroup_complete(group);
4456 ofoperation_get_victim(struct ofoperation *op)
4458 assert(op->type == OFOPERATION_ADD);
4463 pick_datapath_id(const struct ofproto *ofproto)
4465 const struct ofport *port;
4467 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4469 uint8_t ea[ETH_ADDR_LEN];
4472 error = netdev_get_etheraddr(port->netdev, ea);
4474 return eth_addr_to_uint64(ea);
4476 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4477 ofproto->name, netdev_get_name(port->netdev),
4480 return ofproto->fallback_dpid;
4484 pick_fallback_dpid(void)
4486 uint8_t ea[ETH_ADDR_LEN];
4487 eth_addr_nicira_random(ea);
4488 return eth_addr_to_uint64(ea);
4491 /* Table overflow policy. */
4493 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4494 * is not configured to evict rules or if the table contains no evictable
4495 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4497 static struct rule *
4498 choose_rule_to_evict(struct oftable *table)
4500 struct eviction_group *evg;
4502 if (!table->eviction_fields) {
4506 /* In the common case, the outer and inner loops here will each be entered
4509 * - The inner loop normally "return"s in its first iteration. If the
4510 * eviction group has any evictable rules, then it always returns in
4513 * - The outer loop only iterates more than once if the largest eviction
4514 * group has no evictable rules.
4516 * - The outer loop can exit only if table's 'max_flows' is all filled up
4517 * by unevictable rules'. */
4518 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4521 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4522 if (rule->evictable) {
4531 /* Searches 'ofproto' for tables that have more flows than their configured
4532 * maximum and that have flow eviction enabled, and evicts as many flows as
4533 * necessary and currently feasible from them.
4535 * This triggers only when an OpenFlow table has N flows in it and then the
4536 * client configures a maximum number of flows less than N. */
4538 ofproto_evict(struct ofproto *ofproto)
4540 struct ofopgroup *group;
4541 struct oftable *table;
4543 group = ofopgroup_create_unattached(ofproto);
4544 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4545 while (classifier_count(&table->cls) > table->max_flows
4546 && table->eviction_fields) {
4549 rule = choose_rule_to_evict(table);
4550 if (!rule || rule->pending) {
4554 ofoperation_create(group, rule,
4555 OFOPERATION_DELETE, OFPRR_EVICTION);
4556 oftable_remove_rule(rule);
4557 ofproto->ofproto_class->rule_destruct(rule);
4560 ofopgroup_submit(group);
4563 /* Eviction groups. */
4565 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4566 * rules. The priority contains low-order random bits to ensure that eviction
4567 * groups with the same number of rules are prioritized randomly. */
4569 eviction_group_priority(size_t n_rules)
4571 uint16_t size = MIN(UINT16_MAX, n_rules);
4572 return (size << 16) | random_uint16();
4575 /* Updates 'evg', an eviction_group within 'table', following a change that
4576 * adds or removes rules in 'evg'. */
4578 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4580 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4581 eviction_group_priority(heap_count(&evg->rules)));
4584 /* Destroys 'evg', an eviction_group within 'table':
4586 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4587 * rules themselves, just removes them from the eviction group.)
4589 * - Removes 'evg' from 'table'.
4593 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4595 while (!heap_is_empty(&evg->rules)) {
4598 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4599 rule->eviction_group = NULL;
4601 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4602 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4603 heap_destroy(&evg->rules);
4607 /* Removes 'rule' from its eviction group, if any. */
4609 eviction_group_remove_rule(struct rule *rule)
4611 if (rule->eviction_group) {
4612 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4613 struct eviction_group *evg = rule->eviction_group;
4615 rule->eviction_group = NULL;
4616 heap_remove(&evg->rules, &rule->evg_node);
4617 if (heap_is_empty(&evg->rules)) {
4618 eviction_group_destroy(table, evg);
4620 eviction_group_resized(table, evg);
4625 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4626 * returns the hash value. */
4628 eviction_group_hash_rule(struct rule *rule)
4630 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4631 const struct mf_subfield *sf;
4635 hash = table->eviction_group_id_basis;
4636 miniflow_expand(&rule->cr.match.flow, &flow);
4637 for (sf = table->eviction_fields;
4638 sf < &table->eviction_fields[table->n_eviction_fields];
4641 if (mf_are_prereqs_ok(sf->field, &flow)) {
4642 union mf_value value;
4644 mf_get_value(sf->field, &flow, &value);
4646 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4648 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4649 unsigned int start = sf->ofs + sf->n_bits;
4650 bitwise_zero(&value, sf->field->n_bytes, start,
4651 sf->field->n_bytes * 8 - start);
4653 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4655 hash = hash_int(hash, 0);
4662 /* Returns an eviction group within 'table' with the given 'id', creating one
4664 static struct eviction_group *
4665 eviction_group_find(struct oftable *table, uint32_t id)
4667 struct eviction_group *evg;
4669 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4673 evg = xmalloc(sizeof *evg);
4674 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4675 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4676 eviction_group_priority(0));
4677 heap_init(&evg->rules);
4682 /* Returns an eviction priority for 'rule'. The return value should be
4683 * interpreted so that higher priorities make a rule more attractive candidates
4686 rule_eviction_priority(struct rule *rule)
4688 long long int hard_expiration;
4689 long long int idle_expiration;
4690 long long int expiration;
4691 uint32_t expiration_offset;
4693 /* Calculate time of expiration. */
4694 hard_expiration = (rule->hard_timeout
4695 ? rule->modified + rule->hard_timeout * 1000
4697 idle_expiration = (rule->idle_timeout
4698 ? rule->used + rule->idle_timeout * 1000
4700 expiration = MIN(hard_expiration, idle_expiration);
4701 if (expiration == LLONG_MAX) {
4705 /* Calculate the time of expiration as a number of (approximate) seconds
4706 * after program startup.
4708 * This should work OK for program runs that last UINT32_MAX seconds or
4709 * less. Therefore, please restart OVS at least once every 136 years. */
4710 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4712 /* Invert the expiration offset because we're using a max-heap. */
4713 return UINT32_MAX - expiration_offset;
4716 /* Adds 'rule' to an appropriate eviction group for its oftable's
4717 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4718 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4721 * The caller must ensure that 'rule' is not already in an eviction group. */
4723 eviction_group_add_rule(struct rule *rule)
4725 struct ofproto *ofproto = rule->ofproto;
4726 struct oftable *table = &ofproto->tables[rule->table_id];
4728 if (table->eviction_fields
4729 && (rule->hard_timeout || rule->idle_timeout)) {
4730 struct eviction_group *evg;
4732 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4734 rule->eviction_group = evg;
4735 heap_insert(&evg->rules, &rule->evg_node,
4736 rule_eviction_priority(rule));
4737 eviction_group_resized(table, evg);
4743 /* Initializes 'table'. */
4745 oftable_init(struct oftable *table)
4747 memset(table, 0, sizeof *table);
4748 classifier_init(&table->cls);
4749 table->max_flows = UINT_MAX;
4752 /* Destroys 'table', including its classifier and eviction groups.
4754 * The caller is responsible for freeing 'table' itself. */
4756 oftable_destroy(struct oftable *table)
4758 assert(classifier_is_empty(&table->cls));
4759 oftable_disable_eviction(table);
4760 classifier_destroy(&table->cls);
4764 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4765 * string, then 'table' will use its default name.
4767 * This only affects the name exposed for a table exposed through the OpenFlow
4768 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4770 oftable_set_name(struct oftable *table, const char *name)
4772 if (name && name[0]) {
4773 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4774 if (!table->name || strncmp(name, table->name, len)) {
4776 table->name = xmemdup0(name, len);
4784 /* oftables support a choice of two policies when adding a rule would cause the
4785 * number of flows in the table to exceed the configured maximum number: either
4786 * they can refuse to add the new flow or they can evict some existing flow.
4787 * This function configures the former policy on 'table'. */
4789 oftable_disable_eviction(struct oftable *table)
4791 if (table->eviction_fields) {
4792 struct eviction_group *evg, *next;
4794 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4795 &table->eviction_groups_by_id) {
4796 eviction_group_destroy(table, evg);
4798 hmap_destroy(&table->eviction_groups_by_id);
4799 heap_destroy(&table->eviction_groups_by_size);
4801 free(table->eviction_fields);
4802 table->eviction_fields = NULL;
4803 table->n_eviction_fields = 0;
4807 /* oftables support a choice of two policies when adding a rule would cause the
4808 * number of flows in the table to exceed the configured maximum number: either
4809 * they can refuse to add the new flow or they can evict some existing flow.
4810 * This function configures the latter policy on 'table', with fairness based
4811 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4812 * 'n_fields' as 0 disables fairness.) */
4814 oftable_enable_eviction(struct oftable *table,
4815 const struct mf_subfield *fields, size_t n_fields)
4817 struct cls_cursor cursor;
4820 if (table->eviction_fields
4821 && n_fields == table->n_eviction_fields
4823 || !memcmp(fields, table->eviction_fields,
4824 n_fields * sizeof *fields))) {
4829 oftable_disable_eviction(table);
4831 table->n_eviction_fields = n_fields;
4832 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4834 table->eviction_group_id_basis = random_uint32();
4835 hmap_init(&table->eviction_groups_by_id);
4836 heap_init(&table->eviction_groups_by_size);
4838 cls_cursor_init(&cursor, &table->cls, NULL);
4839 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4840 eviction_group_add_rule(rule);
4844 /* Removes 'rule' from the oftable that contains it. */
4846 oftable_remove_rule(struct rule *rule)
4848 struct ofproto *ofproto = rule->ofproto;
4849 struct oftable *table = &ofproto->tables[rule->table_id];
4851 classifier_remove(&table->cls, &rule->cr);
4852 eviction_group_remove_rule(rule);
4855 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4856 * oftable that has an identical cls_rule. Returns the rule that was removed,
4857 * if any, and otherwise NULL. */
4858 static struct rule *
4859 oftable_replace_rule(struct rule *rule)
4861 struct ofproto *ofproto = rule->ofproto;
4862 struct oftable *table = &ofproto->tables[rule->table_id];
4863 struct rule *victim;
4865 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4867 eviction_group_remove_rule(victim);
4869 eviction_group_add_rule(rule);
4873 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4875 oftable_substitute_rule(struct rule *old, struct rule *new)
4878 oftable_replace_rule(new);
4880 oftable_remove_rule(old);
4884 /* unixctl commands. */
4887 ofproto_lookup(const char *name)
4889 struct ofproto *ofproto;
4891 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4893 if (!strcmp(ofproto->name, name)) {
4901 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4902 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4904 struct ofproto *ofproto;
4908 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4909 ds_put_format(&results, "%s\n", ofproto->name);
4911 unixctl_command_reply(conn, ds_cstr(&results));
4912 ds_destroy(&results);
4916 ofproto_unixctl_init(void)
4918 static bool registered;
4924 unixctl_command_register("ofproto/list", "", 0, 0,
4925 ofproto_unixctl_list, NULL);
4928 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4930 * This is deprecated. It is only for compatibility with broken device drivers
4931 * in old versions of Linux that do not properly support VLANs when VLAN
4932 * devices are not used. When broken device drivers are no longer in
4933 * widespread use, we will delete these interfaces. */
4935 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4936 * (exactly) by an OpenFlow rule in 'ofproto'. */
4938 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4940 const struct oftable *oftable;
4942 free(ofproto->vlan_bitmap);
4943 ofproto->vlan_bitmap = bitmap_allocate(4096);
4944 ofproto->vlans_changed = false;
4946 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4947 const struct cls_table *table;
4949 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4950 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4951 const struct cls_rule *rule;
4953 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4954 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4955 bitmap_set1(vlan_bitmap, vid);
4956 bitmap_set1(ofproto->vlan_bitmap, vid);
4963 /* Returns true if new VLANs have come into use by the flow table since the
4964 * last call to ofproto_get_vlan_usage().
4966 * We don't track when old VLANs stop being used. */
4968 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4970 return ofproto->vlans_changed;
4973 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4974 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4975 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4976 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4977 * then the VLAN device is un-enslaved. */
4979 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4980 uint16_t realdev_ofp_port, int vid)
4982 struct ofport *ofport;
4985 assert(vlandev_ofp_port != realdev_ofp_port);
4987 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4989 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4990 ofproto->name, vlandev_ofp_port);
4994 if (!ofproto->ofproto_class->set_realdev) {
4995 if (!vlandev_ofp_port) {
4998 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5002 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5004 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5005 ofproto->name, vlandev_ofp_port,
5006 netdev_get_name(ofport->netdev), strerror(error));