2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
206 static void calc_duration(long long int start, long long int now,
207 uint32_t *sec, uint32_t *nsec);
210 static uint64_t pick_datapath_id(const struct ofproto *);
211 static uint64_t pick_fallback_dpid(void);
212 static void ofproto_destroy__(struct ofproto *);
213 static void update_mtu(struct ofproto *, struct ofport *);
216 static void ofproto_unixctl_init(void);
218 /* All registered ofproto classes, in probe order. */
219 static const struct ofproto_class **ofproto_classes;
220 static size_t n_ofproto_classes;
221 static size_t allocated_ofproto_classes;
223 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
224 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
226 /* Initial mappings of port to OpenFlow number mappings. */
227 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
231 /* Must be called to initialize the ofproto library.
233 * The caller may pass in 'iface_hints', which contains an shash of
234 * "iface_hint" elements indexed by the interface's name. The provider
235 * may use these hints to describe the startup configuration in order to
236 * reinitialize its state. The caller owns the provided data, so a
237 * provider will make copies of anything required. An ofproto provider
238 * will remove any existing state that is not described by the hint, and
239 * may choose to remove it all. */
241 ofproto_init(const struct shash *iface_hints)
243 struct shash_node *node;
246 ofproto_class_register(&ofproto_dpif_class);
248 /* Make a local copy, since we don't own 'iface_hints' elements. */
249 SHASH_FOR_EACH(node, iface_hints) {
250 const struct iface_hint *orig_hint = node->data;
251 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
252 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
254 new_hint->br_name = xstrdup(orig_hint->br_name);
255 new_hint->br_type = xstrdup(br_type);
256 new_hint->ofp_port = orig_hint->ofp_port;
258 shash_add(&init_ofp_ports, node->name, new_hint);
261 for (i = 0; i < n_ofproto_classes; i++) {
262 ofproto_classes[i]->init(&init_ofp_ports);
266 /* 'type' should be a normalized datapath type, as returned by
267 * ofproto_normalize_type(). Returns the corresponding ofproto_class
268 * structure, or a null pointer if there is none registered for 'type'. */
269 static const struct ofproto_class *
270 ofproto_class_find__(const char *type)
274 for (i = 0; i < n_ofproto_classes; i++) {
275 const struct ofproto_class *class = ofproto_classes[i];
280 class->enumerate_types(&types);
281 found = sset_contains(&types, type);
282 sset_destroy(&types);
288 VLOG_WARN("unknown datapath type %s", type);
292 /* Registers a new ofproto class. After successful registration, new ofprotos
293 * of that type can be created using ofproto_create(). */
295 ofproto_class_register(const struct ofproto_class *new_class)
299 for (i = 0; i < n_ofproto_classes; i++) {
300 if (ofproto_classes[i] == new_class) {
305 if (n_ofproto_classes >= allocated_ofproto_classes) {
306 ofproto_classes = x2nrealloc(ofproto_classes,
307 &allocated_ofproto_classes,
308 sizeof *ofproto_classes);
310 ofproto_classes[n_ofproto_classes++] = new_class;
314 /* Unregisters a datapath provider. 'type' must have been previously
315 * registered and not currently be in use by any ofprotos. After
316 * unregistration new datapaths of that type cannot be opened using
317 * ofproto_create(). */
319 ofproto_class_unregister(const struct ofproto_class *class)
323 for (i = 0; i < n_ofproto_classes; i++) {
324 if (ofproto_classes[i] == class) {
325 for (i++; i < n_ofproto_classes; i++) {
326 ofproto_classes[i - 1] = ofproto_classes[i];
332 VLOG_WARN("attempted to unregister an ofproto class that is not "
337 /* Clears 'types' and enumerates all registered ofproto types into it. The
338 * caller must first initialize the sset. */
340 ofproto_enumerate_types(struct sset *types)
344 for (i = 0; i < n_ofproto_classes; i++) {
345 ofproto_classes[i]->enumerate_types(types);
349 /* Returns the fully spelled out name for the given ofproto 'type'.
351 * Normalized type string can be compared with strcmp(). Unnormalized type
352 * string might be the same even if they have different spellings. */
354 ofproto_normalize_type(const char *type)
356 return type && type[0] ? type : "system";
359 /* Clears 'names' and enumerates the names of all known created ofprotos with
360 * the given 'type'. The caller must first initialize the sset. Returns 0 if
361 * successful, otherwise a positive errno value.
363 * Some kinds of datapaths might not be practically enumerable. This is not
364 * considered an error. */
366 ofproto_enumerate_names(const char *type, struct sset *names)
368 const struct ofproto_class *class = ofproto_class_find__(type);
369 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
373 ofproto_create(const char *datapath_name, const char *datapath_type,
374 struct ofproto **ofprotop)
376 const struct ofproto_class *class;
377 struct ofproto *ofproto;
383 ofproto_unixctl_init();
385 datapath_type = ofproto_normalize_type(datapath_type);
386 class = ofproto_class_find__(datapath_type);
388 VLOG_WARN("could not create datapath %s of unknown type %s",
389 datapath_name, datapath_type);
393 ofproto = class->alloc();
395 VLOG_ERR("failed to allocate datapath %s of type %s",
396 datapath_name, datapath_type);
401 memset(ofproto, 0, sizeof *ofproto);
402 ofproto->ofproto_class = class;
403 ofproto->name = xstrdup(datapath_name);
404 ofproto->type = xstrdup(datapath_type);
405 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
406 hash_string(ofproto->name, 0));
407 ofproto->datapath_id = 0;
408 ofproto_set_flow_eviction_threshold(ofproto,
409 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
410 ofproto->forward_bpdu = false;
411 ofproto->fallback_dpid = pick_fallback_dpid();
412 ofproto->mfr_desc = NULL;
413 ofproto->hw_desc = NULL;
414 ofproto->sw_desc = NULL;
415 ofproto->serial_desc = NULL;
416 ofproto->dp_desc = NULL;
417 ofproto->frag_handling = OFPC_FRAG_NORMAL;
418 hmap_init(&ofproto->ports);
419 shash_init(&ofproto->port_by_name);
420 simap_init(&ofproto->ofp_requests);
421 ofproto->max_ports = OFPP_MAX;
422 ofproto->tables = NULL;
423 ofproto->n_tables = 0;
424 list_init(&ofproto->expirable);
425 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
426 ofproto->state = S_OPENFLOW;
427 list_init(&ofproto->pending);
428 ofproto->n_pending = 0;
429 hmap_init(&ofproto->deletions);
430 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
431 ofproto->first_op = ofproto->last_op = LLONG_MIN;
432 ofproto->next_op_report = LLONG_MAX;
433 ofproto->op_backoff = LLONG_MIN;
434 ofproto->vlan_bitmap = NULL;
435 ofproto->vlans_changed = false;
436 ofproto->min_mtu = INT_MAX;
438 error = ofproto->ofproto_class->construct(ofproto);
440 VLOG_ERR("failed to open datapath %s: %s",
441 datapath_name, strerror(error));
442 ofproto_destroy__(ofproto);
446 /* The "max_ports" member should have been set by ->construct(ofproto).
447 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
448 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
449 bitmap_set1(ofproto->ofp_port_ids, 0);
451 /* Check that hidden tables, if any, are at the end. */
452 ovs_assert(ofproto->n_tables);
453 for (i = 0; i + 1 < ofproto->n_tables; i++) {
454 enum oftable_flags flags = ofproto->tables[i].flags;
455 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
457 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
460 ofproto->datapath_id = pick_datapath_id(ofproto);
467 /* Must be called (only) by an ofproto implementation in its constructor
468 * function. See the large comment on 'construct' in struct ofproto_class for
471 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
473 struct oftable *table;
475 ovs_assert(!ofproto->n_tables);
476 ovs_assert(n_tables >= 1 && n_tables <= 255);
478 ofproto->n_tables = n_tables;
479 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
480 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
485 /* To be optionally called (only) by an ofproto implementation in its
486 * constructor function. See the large comment on 'construct' in struct
487 * ofproto_class for details.
489 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
490 * will then ensure that actions passed into the ofproto implementation will
491 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
492 * function is not called, there will be no such restriction.
494 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
495 * the 'max_ports' restriction. */
497 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
499 ovs_assert(max_ports <= OFPP_MAX);
500 ofproto->max_ports = max_ports;
504 ofproto_get_datapath_id(const struct ofproto *ofproto)
506 return ofproto->datapath_id;
510 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
512 uint64_t old_dpid = p->datapath_id;
513 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
514 if (p->datapath_id != old_dpid) {
515 /* Force all active connections to reconnect, since there is no way to
516 * notify a controller that the datapath ID has changed. */
517 ofproto_reconnect_controllers(p);
522 ofproto_set_controllers(struct ofproto *p,
523 const struct ofproto_controller *controllers,
524 size_t n_controllers, uint32_t allowed_versions)
526 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
531 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
533 connmgr_set_fail_mode(p->connmgr, fail_mode);
536 /* Drops the connections between 'ofproto' and all of its controllers, forcing
537 * them to reconnect. */
539 ofproto_reconnect_controllers(struct ofproto *ofproto)
541 connmgr_reconnect(ofproto->connmgr);
544 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
545 * in-band control should guarantee access, in the same way that in-band
546 * control guarantees access to OpenFlow controllers. */
548 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
549 const struct sockaddr_in *extras, size_t n)
551 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
554 /* Sets the OpenFlow queue used by flows set up by in-band control on
555 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
556 * flows will use the default queue. */
558 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
560 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
563 /* Sets the number of flows at which eviction from the kernel flow table
566 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
568 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
569 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
571 ofproto->flow_eviction_threshold = threshold;
575 /* If forward_bpdu is true, the NORMAL action will forward frames with
576 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
577 * the NORMAL action will drop these frames. */
579 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
581 bool old_val = ofproto->forward_bpdu;
582 ofproto->forward_bpdu = forward_bpdu;
583 if (old_val != ofproto->forward_bpdu) {
584 if (ofproto->ofproto_class->forward_bpdu_changed) {
585 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
590 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
591 * 'idle_time', in seconds, and the maximum number of MAC table entries to
594 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
597 if (ofproto->ofproto_class->set_mac_table_config) {
598 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
604 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
607 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
611 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
613 return connmgr_set_snoops(ofproto->connmgr, snoops);
617 ofproto_set_netflow(struct ofproto *ofproto,
618 const struct netflow_options *nf_options)
620 if (nf_options && sset_is_empty(&nf_options->collectors)) {
624 if (ofproto->ofproto_class->set_netflow) {
625 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
627 return nf_options ? EOPNOTSUPP : 0;
632 ofproto_set_sflow(struct ofproto *ofproto,
633 const struct ofproto_sflow_options *oso)
635 if (oso && sset_is_empty(&oso->targets)) {
639 if (ofproto->ofproto_class->set_sflow) {
640 return ofproto->ofproto_class->set_sflow(ofproto, oso);
642 return oso ? EOPNOTSUPP : 0;
647 ofproto_set_ipfix(struct ofproto *ofproto,
648 const struct ofproto_ipfix_bridge_exporter_options *bo,
649 const struct ofproto_ipfix_flow_exporter_options *fo,
652 if (ofproto->ofproto_class->set_ipfix) {
653 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
655 return (bo || fo) ? EOPNOTSUPP : 0;
659 /* Spanning Tree Protocol (STP) configuration. */
661 /* Configures STP on 'ofproto' using the settings defined in 's'. If
662 * 's' is NULL, disables STP.
664 * Returns 0 if successful, otherwise a positive errno value. */
666 ofproto_set_stp(struct ofproto *ofproto,
667 const struct ofproto_stp_settings *s)
669 return (ofproto->ofproto_class->set_stp
670 ? ofproto->ofproto_class->set_stp(ofproto, s)
674 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
675 * 'enabled' member of 's' is false, then the other members are not
678 * Returns 0 if successful, otherwise a positive errno value. */
680 ofproto_get_stp_status(struct ofproto *ofproto,
681 struct ofproto_stp_status *s)
683 return (ofproto->ofproto_class->get_stp_status
684 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
688 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
689 * in 's'. The caller is responsible for assigning STP port numbers
690 * (using the 'port_num' member in the range of 1 through 255, inclusive)
691 * and ensuring there are no duplicates. If the 's' is NULL, then STP
692 * is disabled on the port.
694 * Returns 0 if successful, otherwise a positive errno value.*/
696 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
697 const struct ofproto_port_stp_settings *s)
699 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
701 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
702 ofproto->name, ofp_port);
706 return (ofproto->ofproto_class->set_stp_port
707 ? ofproto->ofproto_class->set_stp_port(ofport, s)
711 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
712 * 's'. If the 'enabled' member in 's' is false, then the other members
713 * are not meaningful.
715 * Returns 0 if successful, otherwise a positive errno value.*/
717 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
718 struct ofproto_port_stp_status *s)
720 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
722 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
723 "port %"PRIu16, ofproto->name, ofp_port);
727 return (ofproto->ofproto_class->get_stp_port_status
728 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
732 /* Queue DSCP configuration. */
734 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
735 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
736 * to implement QoS. Instead, it is used to implement features which require
737 * knowledge of what queues exist on a port, and some basic information about
740 * Returns 0 if successful, otherwise a positive errno value. */
742 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
743 const struct ofproto_port_queue *queues,
746 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
749 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
750 ofproto->name, ofp_port);
754 return (ofproto->ofproto_class->set_queues
755 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
759 /* Connectivity Fault Management configuration. */
761 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
763 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
765 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
766 if (ofport && ofproto->ofproto_class->set_cfm) {
767 ofproto->ofproto_class->set_cfm(ofport, NULL);
771 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
772 * basic configuration from the configuration members in 'cfm', and the remote
773 * maintenance point ID from remote_mpid. Ignores the statistics members of
776 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
778 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
779 const struct cfm_settings *s)
781 struct ofport *ofport;
784 ofport = ofproto_get_port(ofproto, ofp_port);
786 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
787 ofproto->name, ofp_port);
791 /* XXX: For configuration simplicity, we only support one remote_mpid
792 * outside of the CFM module. It's not clear if this is the correct long
793 * term solution or not. */
794 error = (ofproto->ofproto_class->set_cfm
795 ? ofproto->ofproto_class->set_cfm(ofport, s)
798 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
799 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
804 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
805 * 'ofproto' does not have a port 'ofp_port'. */
807 ofproto_port_set_bfd(struct ofproto *ofproto, uint16_t ofp_port,
808 const struct smap *cfg)
810 struct ofport *ofport;
813 ofport = ofproto_get_port(ofproto, ofp_port);
815 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
816 ofproto->name, ofp_port);
820 error = (ofproto->ofproto_class->set_bfd
821 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
824 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
825 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
830 /* Populates 'status' with key value pairs indicating the status of the BFD
831 * session on 'ofp_port'. This information is intended to be populated in the
832 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
835 ofproto_port_get_bfd_status(struct ofproto *ofproto, uint16_t ofp_port,
838 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
839 return (ofport && ofproto->ofproto_class->get_bfd_status
840 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
844 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
845 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
846 * 0 if LACP partner information is not current (generally indicating a
847 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
849 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
851 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
852 return (ofport && ofproto->ofproto_class->port_is_lacp_current
853 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
859 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
860 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
861 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
862 * configuration plus, if there is more than one slave, a bonding
865 * If 'aux' is already registered then this function updates its configuration
866 * to 's'. Otherwise, this function registers a new bundle.
868 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
871 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
872 const struct ofproto_bundle_settings *s)
874 return (ofproto->ofproto_class->bundle_set
875 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
879 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
880 * If no such bundle has been registered, this has no effect. */
882 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
884 return ofproto_bundle_register(ofproto, aux, NULL);
888 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
889 * If 'aux' is already registered then this function updates its configuration
890 * to 's'. Otherwise, this function registers a new mirror. */
892 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
893 const struct ofproto_mirror_settings *s)
895 return (ofproto->ofproto_class->mirror_set
896 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
900 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
901 * If no mirror has been registered, this has no effect. */
903 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
905 return ofproto_mirror_register(ofproto, aux, NULL);
908 /* Retrieves statistics from mirror associated with client data pointer
909 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
910 * 'bytes', respectively. If a particular counters is not supported,
911 * the appropriate argument is set to UINT64_MAX. */
913 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
914 uint64_t *packets, uint64_t *bytes)
916 if (!ofproto->ofproto_class->mirror_get_stats) {
917 *packets = *bytes = UINT64_MAX;
921 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
925 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
926 * which all packets are flooded, instead of using MAC learning. If
927 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
929 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
932 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
934 return (ofproto->ofproto_class->set_flood_vlans
935 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
939 /* Returns true if 'aux' is a registered bundle that is currently in use as the
940 * output for a mirror. */
942 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
944 return (ofproto->ofproto_class->is_mirror_output_bundle
945 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
949 /* Configuration of OpenFlow tables. */
951 /* Returns the number of OpenFlow tables in 'ofproto'. */
953 ofproto_get_n_tables(const struct ofproto *ofproto)
955 return ofproto->n_tables;
958 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
959 * settings from 's'. 'table_id' must be in the range 0 through the number of
960 * OpenFlow tables in 'ofproto' minus 1, inclusive.
962 * For read-only tables, only the name may be configured. */
964 ofproto_configure_table(struct ofproto *ofproto, int table_id,
965 const struct ofproto_table_settings *s)
967 struct oftable *table;
969 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
970 table = &ofproto->tables[table_id];
972 oftable_set_name(table, s->name);
974 if (table->flags & OFTABLE_READONLY) {
979 oftable_enable_eviction(table, s->groups, s->n_groups);
981 oftable_disable_eviction(table);
984 table->max_flows = s->max_flows;
985 if (classifier_count(&table->cls) > table->max_flows
986 && table->eviction_fields) {
987 /* 'table' contains more flows than allowed. We might not be able to
988 * evict them right away because of the asynchronous nature of flow
989 * table changes. Schedule eviction for later. */
990 switch (ofproto->state) {
992 ofproto->state = S_EVICT;
996 /* We're already deleting flows, nothing more to do. */
1003 ofproto_has_snoops(const struct ofproto *ofproto)
1005 return connmgr_has_snoops(ofproto->connmgr);
1009 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1011 connmgr_get_snoops(ofproto->connmgr, snoops);
1015 ofproto_flush__(struct ofproto *ofproto)
1017 struct ofopgroup *group;
1018 struct oftable *table;
1020 if (ofproto->ofproto_class->flush) {
1021 ofproto->ofproto_class->flush(ofproto);
1024 group = ofopgroup_create_unattached(ofproto);
1025 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1026 struct rule *rule, *next_rule;
1027 struct cls_cursor cursor;
1029 if (table->flags & OFTABLE_HIDDEN) {
1033 cls_cursor_init(&cursor, &table->cls, NULL);
1034 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1035 if (!rule->pending) {
1036 ofoperation_create(group, rule, OFOPERATION_DELETE,
1038 oftable_remove_rule(rule);
1039 ofproto->ofproto_class->rule_destruct(rule);
1043 ofopgroup_submit(group);
1047 ofproto_destroy__(struct ofproto *ofproto)
1049 struct oftable *table;
1051 ovs_assert(list_is_empty(&ofproto->pending));
1052 ovs_assert(!ofproto->n_pending);
1054 connmgr_destroy(ofproto->connmgr);
1056 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1057 free(ofproto->name);
1058 free(ofproto->type);
1059 free(ofproto->mfr_desc);
1060 free(ofproto->hw_desc);
1061 free(ofproto->sw_desc);
1062 free(ofproto->serial_desc);
1063 free(ofproto->dp_desc);
1064 hmap_destroy(&ofproto->ports);
1065 shash_destroy(&ofproto->port_by_name);
1066 bitmap_free(ofproto->ofp_port_ids);
1067 simap_destroy(&ofproto->ofp_requests);
1069 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1070 oftable_destroy(table);
1072 free(ofproto->tables);
1074 hmap_destroy(&ofproto->deletions);
1076 free(ofproto->vlan_bitmap);
1078 ofproto->ofproto_class->dealloc(ofproto);
1082 ofproto_destroy(struct ofproto *p)
1084 struct ofport *ofport, *next_ofport;
1091 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1092 ofport_destroy(ofport);
1095 p->ofproto_class->destruct(p);
1096 ofproto_destroy__(p);
1099 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1100 * kernel datapath, for example, this destroys the datapath in the kernel, and
1101 * with the netdev-based datapath, it tears down the data structures that
1102 * represent the datapath.
1104 * The datapath should not be currently open as an ofproto. */
1106 ofproto_delete(const char *name, const char *type)
1108 const struct ofproto_class *class = ofproto_class_find__(type);
1109 return (!class ? EAFNOSUPPORT
1110 : !class->del ? EACCES
1111 : class->del(type, name));
1115 process_port_change(struct ofproto *ofproto, int error, char *devname)
1117 if (error == ENOBUFS) {
1118 reinit_ports(ofproto);
1119 } else if (!error) {
1120 update_port(ofproto, devname);
1126 ofproto_type_run(const char *datapath_type)
1128 const struct ofproto_class *class;
1131 datapath_type = ofproto_normalize_type(datapath_type);
1132 class = ofproto_class_find__(datapath_type);
1134 error = class->type_run ? class->type_run(datapath_type) : 0;
1135 if (error && error != EAGAIN) {
1136 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1137 datapath_type, strerror(error));
1143 ofproto_type_run_fast(const char *datapath_type)
1145 const struct ofproto_class *class;
1148 datapath_type = ofproto_normalize_type(datapath_type);
1149 class = ofproto_class_find__(datapath_type);
1151 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1152 if (error && error != EAGAIN) {
1153 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1154 datapath_type, strerror(error));
1160 ofproto_type_wait(const char *datapath_type)
1162 const struct ofproto_class *class;
1164 datapath_type = ofproto_normalize_type(datapath_type);
1165 class = ofproto_class_find__(datapath_type);
1167 if (class->type_wait) {
1168 class->type_wait(datapath_type);
1173 ofproto_run(struct ofproto *p)
1175 struct sset changed_netdevs;
1176 const char *changed_netdev;
1177 struct ofport *ofport;
1180 error = p->ofproto_class->run(p);
1181 if (error && error != EAGAIN) {
1182 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1185 if (p->ofproto_class->port_poll) {
1188 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1189 process_port_change(p, error, devname);
1193 /* Update OpenFlow port status for any port whose netdev has changed.
1195 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1196 * destroyed, so it's not safe to update ports directly from the
1197 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1198 * need this two-phase approach. */
1199 sset_init(&changed_netdevs);
1200 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1201 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1202 if (ofport->change_seq != change_seq) {
1203 ofport->change_seq = change_seq;
1204 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1207 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1208 update_port(p, changed_netdev);
1210 sset_destroy(&changed_netdevs);
1214 connmgr_run(p->connmgr, handle_openflow);
1218 connmgr_run(p->connmgr, NULL);
1220 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1221 p->state = S_OPENFLOW;
1226 connmgr_run(p->connmgr, NULL);
1228 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1229 connmgr_flushed(p->connmgr);
1230 p->state = S_OPENFLOW;
1238 if (time_msec() >= p->next_op_report) {
1239 long long int ago = (time_msec() - p->first_op) / 1000;
1240 long long int interval = (p->last_op - p->first_op) / 1000;
1244 ds_put_format(&s, "%d flow_mods ",
1245 p->n_add + p->n_delete + p->n_modify);
1246 if (interval == ago) {
1247 ds_put_format(&s, "in the last %lld s", ago);
1248 } else if (interval) {
1249 ds_put_format(&s, "in the %lld s starting %lld s ago",
1252 ds_put_format(&s, "%lld s ago", ago);
1255 ds_put_cstr(&s, " (");
1257 ds_put_format(&s, "%d adds, ", p->n_add);
1260 ds_put_format(&s, "%d deletes, ", p->n_delete);
1263 ds_put_format(&s, "%d modifications, ", p->n_modify);
1266 ds_put_char(&s, ')');
1268 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1271 p->n_add = p->n_delete = p->n_modify = 0;
1272 p->next_op_report = LLONG_MAX;
1278 /* Performs periodic activity required by 'ofproto' that needs to be done
1279 * with the least possible latency.
1281 * It makes sense to call this function a couple of times per poll loop, to
1282 * provide a significant performance boost on some benchmarks with the
1283 * ofproto-dpif implementation. */
1285 ofproto_run_fast(struct ofproto *p)
1289 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1290 if (error && error != EAGAIN) {
1291 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1292 p->name, strerror(error));
1298 ofproto_wait(struct ofproto *p)
1300 struct ofport *ofport;
1302 p->ofproto_class->wait(p);
1303 if (p->ofproto_class->port_poll_wait) {
1304 p->ofproto_class->port_poll_wait(p);
1307 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1308 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1309 poll_immediate_wake();
1315 connmgr_wait(p->connmgr, true);
1320 connmgr_wait(p->connmgr, false);
1321 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1322 poll_immediate_wake();
1329 ofproto_is_alive(const struct ofproto *p)
1331 return connmgr_has_controllers(p->connmgr);
1334 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1335 * memory_report(). */
1337 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1339 const struct oftable *table;
1340 unsigned int n_rules;
1342 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1343 simap_increase(usage, "ops",
1344 ofproto->n_pending + hmap_count(&ofproto->deletions));
1347 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1348 n_rules += classifier_count(&table->cls);
1350 simap_increase(usage, "rules", n_rules);
1352 if (ofproto->ofproto_class->get_memory_usage) {
1353 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1356 connmgr_get_memory_usage(ofproto->connmgr, usage);
1360 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1363 connmgr_get_controller_info(ofproto->connmgr, info);
1367 ofproto_free_ofproto_controller_info(struct shash *info)
1369 connmgr_free_controller_info(info);
1372 /* Makes a deep copy of 'old' into 'port'. */
1374 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1376 port->name = xstrdup(old->name);
1377 port->type = xstrdup(old->type);
1378 port->ofp_port = old->ofp_port;
1381 /* Frees memory allocated to members of 'ofproto_port'.
1383 * Do not call this function on an ofproto_port obtained from
1384 * ofproto_port_dump_next(): that function retains ownership of the data in the
1387 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1389 free(ofproto_port->name);
1390 free(ofproto_port->type);
1393 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1395 * This function provides no status indication. An error status for the entire
1396 * dump operation is provided when it is completed by calling
1397 * ofproto_port_dump_done().
1400 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1401 const struct ofproto *ofproto)
1403 dump->ofproto = ofproto;
1404 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1408 /* Attempts to retrieve another port from 'dump', which must have been created
1409 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1410 * 'port' and returns true. On failure, returns false.
1412 * Failure might indicate an actual error or merely that the last port has been
1413 * dumped. An error status for the entire dump operation is provided when it
1414 * is completed by calling ofproto_port_dump_done().
1416 * The ofproto owns the data stored in 'port'. It will remain valid until at
1417 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1418 * ofproto_port_dump_done(). */
1420 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1421 struct ofproto_port *port)
1423 const struct ofproto *ofproto = dump->ofproto;
1429 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1432 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1438 /* Completes port table dump operation 'dump', which must have been created
1439 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1440 * error-free, otherwise a positive errno value describing the problem. */
1442 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1444 const struct ofproto *ofproto = dump->ofproto;
1446 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1449 return dump->error == EOF ? 0 : dump->error;
1452 /* Returns the type to pass to netdev_open() when a datapath of type
1453 * 'datapath_type' has a port of type 'port_type', for a few special
1454 * cases when a netdev type differs from a port type. For example, when
1455 * using the userspace datapath, a port of type "internal" needs to be
1458 * Returns either 'type' itself or a string literal, which must not be
1461 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1463 const struct ofproto_class *class;
1465 datapath_type = ofproto_normalize_type(datapath_type);
1466 class = ofproto_class_find__(datapath_type);
1471 return (class->port_open_type
1472 ? class->port_open_type(datapath_type, port_type)
1476 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1477 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1478 * the port's OpenFlow port number.
1480 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1481 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1482 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1483 * 'ofp_portp' is non-null). */
1485 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1486 uint16_t *ofp_portp)
1488 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1491 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1493 const char *netdev_name = netdev_get_name(netdev);
1495 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1496 update_port(ofproto, netdev_name);
1499 struct ofproto_port ofproto_port;
1501 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1503 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1504 ofproto_port_destroy(&ofproto_port);
1509 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1510 * initializes '*port' appropriately; on failure, returns a positive errno
1513 * The caller owns the data in 'ofproto_port' and must free it with
1514 * ofproto_port_destroy() when it is no longer needed. */
1516 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1517 struct ofproto_port *port)
1521 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1523 memset(port, 0, sizeof *port);
1528 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1529 * Returns 0 if successful, otherwise a positive errno. */
1531 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1533 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1534 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1535 struct simap_node *ofp_request_node;
1538 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1539 if (ofp_request_node) {
1540 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1543 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1544 if (!error && ofport) {
1545 /* 'name' is the netdev's name and update_port() is going to close the
1546 * netdev. Just in case update_port() refers to 'name' after it
1547 * destroys 'ofport', make a copy of it around the update_port()
1549 char *devname = xstrdup(name);
1550 update_port(ofproto, devname);
1556 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1557 * performs the 'n_actions' actions in 'actions'. The new flow will not
1560 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1561 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1562 * controllers; otherwise, it will be hidden.
1564 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1566 * This is a helper function for in-band control and fail-open. */
1568 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1569 unsigned int priority,
1570 const struct ofpact *ofpacts, size_t ofpacts_len)
1572 const struct rule *rule;
1574 rule = rule_from_cls_rule(classifier_find_match_exactly(
1575 &ofproto->tables[0].cls, match, priority));
1576 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1577 ofpacts, ofpacts_len)) {
1578 struct ofputil_flow_mod fm;
1580 memset(&fm, 0, sizeof fm);
1582 fm.priority = priority;
1583 fm.buffer_id = UINT32_MAX;
1584 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1585 fm.ofpacts_len = ofpacts_len;
1586 add_flow(ofproto, NULL, &fm, NULL);
1591 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1592 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1593 * operation cannot be initiated now but may be retried later.
1595 * This is a helper function for in-band control and fail-open. */
1597 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1599 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1602 /* Searches for a rule with matching criteria exactly equal to 'target' in
1603 * ofproto's table 0 and, if it finds one, deletes it.
1605 * This is a helper function for in-band control and fail-open. */
1607 ofproto_delete_flow(struct ofproto *ofproto,
1608 const struct match *target, unsigned int priority)
1612 rule = rule_from_cls_rule(classifier_find_match_exactly(
1613 &ofproto->tables[0].cls, target, priority));
1615 /* No such rule -> success. */
1617 } else if (rule->pending) {
1618 /* An operation on the rule is already pending -> failure.
1619 * Caller must retry later if it's important. */
1622 /* Initiate deletion -> success. */
1623 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1624 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1625 oftable_remove_rule(rule);
1626 ofproto->ofproto_class->rule_destruct(rule);
1627 ofopgroup_submit(group);
1633 /* Starts the process of deleting all of the flows from all of ofproto's flow
1634 * tables and then reintroducing the flows required by in-band control and
1635 * fail-open. The process will complete in a later call to ofproto_run(). */
1637 ofproto_flush_flows(struct ofproto *ofproto)
1639 COVERAGE_INC(ofproto_flush);
1640 ofproto->state = S_FLUSH;
1644 reinit_ports(struct ofproto *p)
1646 struct ofproto_port_dump dump;
1647 struct sset devnames;
1648 struct ofport *ofport;
1649 struct ofproto_port ofproto_port;
1650 const char *devname;
1652 COVERAGE_INC(ofproto_reinit_ports);
1654 sset_init(&devnames);
1655 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1656 sset_add(&devnames, netdev_get_name(ofport->netdev));
1658 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1659 sset_add(&devnames, ofproto_port.name);
1662 SSET_FOR_EACH (devname, &devnames) {
1663 update_port(p, devname);
1665 sset_destroy(&devnames);
1669 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1672 uint16_t end_port_no = ofproto->alloc_port_no;
1674 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1675 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1677 if (ofp_port >= ofproto->max_ports
1678 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1679 /* Search for a free OpenFlow port number. We try not to
1680 * immediately reuse them to prevent problems due to old
1683 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1684 ofproto->alloc_port_no = 0;
1686 if (!bitmap_is_set(ofproto->ofp_port_ids,
1687 ofproto->alloc_port_no)) {
1688 ofp_port = ofproto->alloc_port_no;
1691 if (ofproto->alloc_port_no == end_port_no) {
1696 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1701 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1703 if (ofp_port < ofproto->max_ports) {
1704 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1708 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1709 * pointer if the netdev cannot be opened. On success, also fills in
1711 static struct netdev *
1712 ofport_open(struct ofproto *ofproto,
1713 struct ofproto_port *ofproto_port,
1714 struct ofputil_phy_port *pp)
1716 enum netdev_flags flags;
1717 struct netdev *netdev;
1720 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1722 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1723 "cannot be opened (%s)",
1725 ofproto_port->name, ofproto_port->ofp_port,
1726 ofproto_port->name, strerror(error));
1730 if (ofproto_port->ofp_port == OFPP_NONE) {
1731 if (!strcmp(ofproto->name, ofproto_port->name)) {
1732 ofproto_port->ofp_port = OFPP_LOCAL;
1734 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1735 ofproto_port->name);
1738 pp->port_no = ofproto_port->ofp_port;
1739 netdev_get_etheraddr(netdev, pp->hw_addr);
1740 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1741 netdev_get_flags(netdev, &flags);
1742 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1743 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1744 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1745 &pp->supported, &pp->peer);
1746 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1747 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1752 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1753 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1756 ofport_equal(const struct ofputil_phy_port *a,
1757 const struct ofputil_phy_port *b)
1759 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1760 && a->state == b->state
1761 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1762 && a->curr == b->curr
1763 && a->advertised == b->advertised
1764 && a->supported == b->supported
1765 && a->peer == b->peer
1766 && a->curr_speed == b->curr_speed
1767 && a->max_speed == b->max_speed);
1770 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1771 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1772 * one with the same name or port number). */
1774 ofport_install(struct ofproto *p,
1775 struct netdev *netdev, const struct ofputil_phy_port *pp)
1777 const char *netdev_name = netdev_get_name(netdev);
1778 struct ofport *ofport;
1781 /* Create ofport. */
1782 ofport = p->ofproto_class->port_alloc();
1787 ofport->ofproto = p;
1788 ofport->netdev = netdev;
1789 ofport->change_seq = netdev_change_seq(netdev);
1791 ofport->ofp_port = pp->port_no;
1792 ofport->created = time_msec();
1794 /* Add port to 'p'. */
1795 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1796 shash_add(&p->port_by_name, netdev_name, ofport);
1798 update_mtu(p, ofport);
1800 /* Let the ofproto_class initialize its private data. */
1801 error = p->ofproto_class->port_construct(ofport);
1805 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1809 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1810 p->name, netdev_name, strerror(error));
1812 ofport_destroy__(ofport);
1814 netdev_close(netdev);
1818 /* Removes 'ofport' from 'p' and destroys it. */
1820 ofport_remove(struct ofport *ofport)
1822 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1824 ofport_destroy(ofport);
1827 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1830 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1832 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1834 ofport_remove(port);
1838 /* Updates 'port' with new 'pp' description.
1840 * Does not handle a name or port number change. The caller must implement
1841 * such a change as a delete followed by an add. */
1843 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1845 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1846 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1847 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1848 port->pp.state = pp->state;
1849 port->pp.curr = pp->curr;
1850 port->pp.advertised = pp->advertised;
1851 port->pp.supported = pp->supported;
1852 port->pp.peer = pp->peer;
1853 port->pp.curr_speed = pp->curr_speed;
1854 port->pp.max_speed = pp->max_speed;
1856 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1859 /* Update OpenFlow 'state' in 'port' and notify controller. */
1861 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1863 if (port->pp.state != state) {
1864 port->pp.state = state;
1865 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1871 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1873 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1875 if (port->ofproto->ofproto_class->set_realdev) {
1876 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1878 if (port->ofproto->ofproto_class->set_stp_port) {
1879 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1881 if (port->ofproto->ofproto_class->set_cfm) {
1882 port->ofproto->ofproto_class->set_cfm(port, NULL);
1884 if (port->ofproto->ofproto_class->bundle_remove) {
1885 port->ofproto->ofproto_class->bundle_remove(port);
1891 ofport_destroy__(struct ofport *port)
1893 struct ofproto *ofproto = port->ofproto;
1894 const char *name = netdev_get_name(port->netdev);
1896 hmap_remove(&ofproto->ports, &port->hmap_node);
1897 shash_delete(&ofproto->port_by_name,
1898 shash_find(&ofproto->port_by_name, name));
1900 netdev_close(port->netdev);
1901 ofproto->ofproto_class->port_dealloc(port);
1905 ofport_destroy(struct ofport *port)
1908 dealloc_ofp_port(port->ofproto, port->ofp_port);
1909 port->ofproto->ofproto_class->port_destruct(port);
1910 ofport_destroy__(port);
1915 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1917 struct ofport *port;
1919 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1920 hash_int(ofp_port, 0), &ofproto->ports) {
1921 if (port->ofp_port == ofp_port) {
1929 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1931 struct ofproto *ofproto = port->ofproto;
1934 if (ofproto->ofproto_class->port_get_stats) {
1935 error = ofproto->ofproto_class->port_get_stats(port, stats);
1944 update_port(struct ofproto *ofproto, const char *name)
1946 struct ofproto_port ofproto_port;
1947 struct ofputil_phy_port pp;
1948 struct netdev *netdev;
1949 struct ofport *port;
1951 COVERAGE_INC(ofproto_update_port);
1953 /* Fetch 'name''s location and properties from the datapath. */
1954 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1955 ? ofport_open(ofproto, &ofproto_port, &pp)
1958 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1959 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1960 struct netdev *old_netdev = port->netdev;
1962 /* 'name' hasn't changed location. Any properties changed? */
1963 if (!ofport_equal(&port->pp, &pp)) {
1964 ofport_modified(port, &pp);
1967 update_mtu(ofproto, port);
1969 /* Install the newly opened netdev in case it has changed.
1970 * Don't close the old netdev yet in case port_modified has to
1971 * remove a retained reference to it.*/
1972 port->netdev = netdev;
1973 port->change_seq = netdev_change_seq(netdev);
1975 if (port->ofproto->ofproto_class->port_modified) {
1976 port->ofproto->ofproto_class->port_modified(port);
1979 netdev_close(old_netdev);
1981 /* If 'port' is nonnull then its name differs from 'name' and thus
1982 * we should delete it. If we think there's a port named 'name'
1983 * then its port number must be wrong now so delete it too. */
1985 ofport_remove(port);
1987 ofport_remove_with_name(ofproto, name);
1988 ofport_install(ofproto, netdev, &pp);
1991 /* Any port named 'name' is gone now. */
1992 ofport_remove_with_name(ofproto, name);
1994 ofproto_port_destroy(&ofproto_port);
1998 init_ports(struct ofproto *p)
2000 struct ofproto_port_dump dump;
2001 struct ofproto_port ofproto_port;
2002 struct shash_node *node, *next;
2004 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2005 const char *name = ofproto_port.name;
2007 if (shash_find(&p->port_by_name, name)) {
2008 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2011 struct ofputil_phy_port pp;
2012 struct netdev *netdev;
2014 /* Check if an OpenFlow port number had been requested. */
2015 node = shash_find(&init_ofp_ports, name);
2017 const struct iface_hint *iface_hint = node->data;
2018 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
2021 netdev = ofport_open(p, &ofproto_port, &pp);
2023 ofport_install(p, netdev, &pp);
2028 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2029 struct iface_hint *iface_hint = node->data;
2031 if (!strcmp(iface_hint->br_name, p->name)) {
2032 free(iface_hint->br_name);
2033 free(iface_hint->br_type);
2035 shash_delete(&init_ofp_ports, node);
2042 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2043 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2045 find_min_mtu(struct ofproto *p)
2047 struct ofport *ofport;
2050 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2051 struct netdev *netdev = ofport->netdev;
2054 /* Skip any internal ports, since that's what we're trying to
2056 if (!strcmp(netdev_get_type(netdev), "internal")) {
2060 if (netdev_get_mtu(netdev, &dev_mtu)) {
2063 if (!mtu || dev_mtu < mtu) {
2068 return mtu ? mtu: ETH_PAYLOAD_MAX;
2071 /* Update MTU of all datapath devices on 'p' to the minimum of the
2072 * non-datapath ports in event of 'port' added or changed. */
2074 update_mtu(struct ofproto *p, struct ofport *port)
2076 struct ofport *ofport;
2077 struct netdev *netdev = port->netdev;
2078 int dev_mtu, old_min;
2080 if (netdev_get_mtu(netdev, &dev_mtu)) {
2084 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2085 if (dev_mtu > p->min_mtu) {
2086 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2087 dev_mtu = p->min_mtu;
2090 port->mtu = dev_mtu;
2094 /* For non-internal port find new min mtu. */
2095 old_min = p->min_mtu;
2096 port->mtu = dev_mtu;
2097 p->min_mtu = find_min_mtu(p);
2098 if (p->min_mtu == old_min) {
2102 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2103 struct netdev *netdev = ofport->netdev;
2105 if (!strcmp(netdev_get_type(netdev), "internal")) {
2106 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2107 ofport->mtu = p->min_mtu;
2114 ofproto_rule_destroy__(struct rule *rule)
2117 cls_rule_destroy(&rule->cr);
2118 free(rule->ofpacts);
2119 rule->ofproto->ofproto_class->rule_dealloc(rule);
2123 /* This function allows an ofproto implementation to destroy any rules that
2124 * remain when its ->destruct() function is called. The caller must have
2125 * already uninitialized any derived members of 'rule' (step 5 described in the
2126 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2127 * This function implements steps 6 and 7.
2129 * This function should only be called from an ofproto implementation's
2130 * ->destruct() function. It is not suitable elsewhere. */
2132 ofproto_rule_destroy(struct rule *rule)
2134 ovs_assert(!rule->pending);
2135 oftable_remove_rule(rule);
2136 ofproto_rule_destroy__(rule);
2139 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2140 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2142 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2144 return (port == OFPP_ANY
2145 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2148 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2149 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2151 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2153 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2158 case OFOPERATION_ADD:
2159 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2161 case OFOPERATION_DELETE:
2164 case OFOPERATION_MODIFY:
2165 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2171 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2172 * statistics appropriately. 'packet' must have at least sizeof(struct
2173 * ofp10_packet_in) bytes of headroom.
2175 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2176 * with statistics for 'packet' either way.
2178 * Takes ownership of 'packet'. */
2180 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2184 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2186 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2187 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2190 /* Returns true if 'rule' should be hidden from the controller.
2192 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2193 * (e.g. by in-band control) and are intentionally hidden from the
2196 ofproto_rule_is_hidden(const struct rule *rule)
2198 return rule->cr.priority > UINT16_MAX;
2201 static enum oftable_flags
2202 rule_get_flags(const struct rule *rule)
2204 return rule->ofproto->tables[rule->table_id].flags;
2208 rule_is_modifiable(const struct rule *rule)
2210 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2214 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2216 ofconn_send_reply(ofconn, make_echo_reply(oh));
2221 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2223 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2224 struct ofputil_switch_features features;
2225 struct ofport *port;
2231 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2233 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2235 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2236 * if present, are always at the end.) */
2237 n_tables = ofproto->n_tables;
2238 for (i = 0; i < ofproto->n_tables; i++) {
2239 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2245 features.datapath_id = ofproto->datapath_id;
2246 features.n_buffers = pktbuf_capacity();
2247 features.n_tables = n_tables;
2248 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2249 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2251 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2253 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2254 features.auxiliary_id = 0;
2255 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2257 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2258 ofputil_put_switch_features_port(&port->pp, b);
2261 ofconn_send_reply(ofconn, b);
2266 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2268 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2269 struct ofp_switch_config *osc;
2270 enum ofp_config_flags flags;
2274 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2275 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2276 flags = ofproto->frag_handling;
2277 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2278 if (oh->version < OFP13_VERSION
2279 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2280 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2282 osc->flags = htons(flags);
2283 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2284 ofconn_send_reply(ofconn, buf);
2290 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2292 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2293 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2294 uint16_t flags = ntohs(osc->flags);
2296 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2297 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2298 enum ofp_config_flags cur = ofproto->frag_handling;
2299 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2301 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2303 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2304 ofproto->frag_handling = next;
2306 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2308 ofputil_frag_handling_to_string(next));
2312 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2313 ofconn_set_invalid_ttl_to_controller(ofconn,
2314 (oh->version < OFP13_VERSION
2315 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2317 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2322 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2323 * error message code for the caller to propagate upward. Otherwise, returns
2326 * The log message mentions 'msg_type'. */
2328 reject_slave_controller(struct ofconn *ofconn)
2330 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2331 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2332 return OFPERR_OFPBRC_EPERM;
2339 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2341 struct ofproto *p = ofconn_get_ofproto(ofconn);
2342 struct ofputil_packet_out po;
2343 struct ofpbuf *payload;
2344 uint64_t ofpacts_stub[1024 / 8];
2345 struct ofpbuf ofpacts;
2349 COVERAGE_INC(ofproto_packet_out);
2351 error = reject_slave_controller(ofconn);
2356 /* Decode message. */
2357 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2358 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2360 goto exit_free_ofpacts;
2362 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2363 error = OFPERR_OFPBRC_BAD_PORT;
2364 goto exit_free_ofpacts;
2369 if (po.buffer_id != UINT32_MAX) {
2370 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2371 if (error || !payload) {
2372 goto exit_free_ofpacts;
2375 payload = xmalloc(sizeof *payload);
2376 ofpbuf_use_const(payload, po.packet, po.packet_len);
2379 /* Verify actions against packet, then send packet if successful. */
2380 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2381 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2383 error = p->ofproto_class->packet_out(p, payload, &flow,
2384 po.ofpacts, po.ofpacts_len);
2386 ofpbuf_delete(payload);
2389 ofpbuf_uninit(&ofpacts);
2395 update_port_config(struct ofport *port,
2396 enum ofputil_port_config config,
2397 enum ofputil_port_config mask)
2399 enum ofputil_port_config old_config = port->pp.config;
2400 enum ofputil_port_config toggle;
2402 toggle = (config ^ port->pp.config) & mask;
2403 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2404 if (config & OFPUTIL_PC_PORT_DOWN) {
2405 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2407 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2409 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2412 port->pp.config ^= toggle;
2413 if (port->pp.config != old_config) {
2414 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2419 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2421 struct ofproto *p = ofconn_get_ofproto(ofconn);
2422 struct ofputil_port_mod pm;
2423 struct ofport *port;
2426 error = reject_slave_controller(ofconn);
2431 error = ofputil_decode_port_mod(oh, &pm);
2436 port = ofproto_get_port(p, pm.port_no);
2438 return OFPERR_OFPPMFC_BAD_PORT;
2439 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2440 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2442 update_port_config(port, pm.config, pm.mask);
2444 netdev_set_advertisements(port->netdev, pm.advertise);
2451 handle_desc_stats_request(struct ofconn *ofconn,
2452 const struct ofp_header *request)
2454 static const char *default_mfr_desc = "Nicira, Inc.";
2455 static const char *default_hw_desc = "Open vSwitch";
2456 static const char *default_sw_desc = VERSION;
2457 static const char *default_serial_desc = "None";
2458 static const char *default_dp_desc = "None";
2460 struct ofproto *p = ofconn_get_ofproto(ofconn);
2461 struct ofp_desc_stats *ods;
2464 msg = ofpraw_alloc_stats_reply(request, 0);
2465 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2466 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2467 sizeof ods->mfr_desc);
2468 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2469 sizeof ods->hw_desc);
2470 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2471 sizeof ods->sw_desc);
2472 ovs_strlcpy(ods->serial_num,
2473 p->serial_desc ? p->serial_desc : default_serial_desc,
2474 sizeof ods->serial_num);
2475 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2476 sizeof ods->dp_desc);
2477 ofconn_send_reply(ofconn, msg);
2483 handle_table_stats_request(struct ofconn *ofconn,
2484 const struct ofp_header *request)
2486 struct ofproto *p = ofconn_get_ofproto(ofconn);
2487 struct ofp12_table_stats *ots;
2492 /* Set up default values.
2494 * ofp12_table_stats is used as a generic structure as
2495 * it is able to hold all the fields for ofp10_table_stats
2496 * and ofp11_table_stats (and of course itself).
2498 ots = xcalloc(p->n_tables, sizeof *ots);
2499 for (i = 0; i < p->n_tables; i++) {
2500 ots[i].table_id = i;
2501 sprintf(ots[i].name, "table%zu", i);
2502 ots[i].match = htonll(OFPXMT12_MASK);
2503 ots[i].wildcards = htonll(OFPXMT12_MASK);
2504 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2505 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2506 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2507 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2508 ots[i].metadata_match = htonll(UINT64_MAX);
2509 ots[i].metadata_write = htonll(UINT64_MAX);
2510 ots[i].instructions = htonl(OFPIT11_ALL);
2511 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2512 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2513 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2516 p->ofproto_class->get_tables(p, ots);
2518 /* Post-process the tables, dropping hidden tables. */
2519 n_tables = p->n_tables;
2520 for (i = 0; i < p->n_tables; i++) {
2521 const struct oftable *table = &p->tables[i];
2523 if (table->flags & OFTABLE_HIDDEN) {
2529 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2532 if (table->max_flows < ntohl(ots[i].max_entries)) {
2533 ots[i].max_entries = htonl(table->max_flows);
2537 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2538 ofconn_send_reply(ofconn, msg);
2546 append_port_stat(struct ofport *port, struct list *replies)
2548 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2550 calc_duration(port->created, time_msec(),
2551 &ops.duration_sec, &ops.duration_nsec);
2553 /* Intentionally ignore return value, since errors will set
2554 * 'stats' to all-1s, which is correct for OpenFlow, and
2555 * netdev_get_stats() will log errors. */
2556 ofproto_port_get_stats(port, &ops.stats);
2558 ofputil_append_port_stat(replies, &ops);
2562 handle_port_stats_request(struct ofconn *ofconn,
2563 const struct ofp_header *request)
2565 struct ofproto *p = ofconn_get_ofproto(ofconn);
2566 struct ofport *port;
2567 struct list replies;
2571 error = ofputil_decode_port_stats_request(request, &port_no);
2576 ofpmp_init(&replies, request);
2577 if (port_no != OFPP_ANY) {
2578 port = ofproto_get_port(p, port_no);
2580 append_port_stat(port, &replies);
2583 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2584 append_port_stat(port, &replies);
2588 ofconn_send_replies(ofconn, &replies);
2593 handle_port_desc_stats_request(struct ofconn *ofconn,
2594 const struct ofp_header *request)
2596 struct ofproto *p = ofconn_get_ofproto(ofconn);
2597 enum ofp_version version;
2598 struct ofport *port;
2599 struct list replies;
2601 ofpmp_init(&replies, request);
2603 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2604 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2605 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2608 ofconn_send_replies(ofconn, &replies);
2613 calc_duration(long long int start, long long int now,
2614 uint32_t *sec, uint32_t *nsec)
2616 long long int msecs = now - start;
2617 *sec = msecs / 1000;
2618 *nsec = (msecs % 1000) * (1000 * 1000);
2621 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2622 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2624 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2626 return (table_id == 0xff || table_id < ofproto->n_tables
2628 : OFPERR_OFPBRC_BAD_TABLE_ID);
2632 static struct oftable *
2633 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2635 struct oftable *table;
2637 for (table = &ofproto->tables[table_id];
2638 table < &ofproto->tables[ofproto->n_tables];
2640 if (!(table->flags & OFTABLE_HIDDEN)) {
2648 static struct oftable *
2649 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2651 if (table_id == 0xff) {
2652 return next_visible_table(ofproto, 0);
2653 } else if (table_id < ofproto->n_tables) {
2654 return &ofproto->tables[table_id];
2660 static struct oftable *
2661 next_matching_table(const struct ofproto *ofproto,
2662 const struct oftable *table, uint8_t table_id)
2664 return (table_id == 0xff
2665 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2669 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2671 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2672 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2674 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2675 * only once, for that table. (This can be used to access tables marked
2678 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2679 * entered at all. (Perhaps you should have validated TABLE_ID with
2680 * check_table_id().)
2682 * All parameters are evaluated multiple times.
2684 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2685 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2687 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2689 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2690 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2691 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2694 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2695 * to 'out_port' are included.
2697 * Hidden rules are always omitted.
2699 * Returns 0 on success, otherwise an OpenFlow error code. */
2701 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2702 const struct match *match,
2703 ovs_be64 cookie, ovs_be64 cookie_mask,
2704 uint16_t out_port, struct list *rules)
2706 struct oftable *table;
2710 error = check_table_id(ofproto, table_id);
2716 cls_rule_init(&cr, match, 0);
2717 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2718 struct cls_cursor cursor;
2721 cls_cursor_init(&cursor, &table->cls, &cr);
2722 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2723 if (rule->pending) {
2724 error = OFPROTO_POSTPONE;
2727 if (!ofproto_rule_is_hidden(rule)
2728 && ofproto_rule_has_out_port(rule, out_port)
2729 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2730 list_push_back(rules, &rule->ofproto_node);
2736 cls_rule_destroy(&cr);
2740 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2741 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2742 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2745 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2746 * to 'out_port' are included.
2748 * Hidden rules are always omitted.
2750 * Returns 0 on success, otherwise an OpenFlow error code. */
2752 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2753 const struct match *match, unsigned int priority,
2754 ovs_be64 cookie, ovs_be64 cookie_mask,
2755 uint16_t out_port, struct list *rules)
2757 struct oftable *table;
2761 error = check_table_id(ofproto, table_id);
2767 cls_rule_init(&cr, match, priority);
2768 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2771 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2774 if (rule->pending) {
2775 error = OFPROTO_POSTPONE;
2778 if (!ofproto_rule_is_hidden(rule)
2779 && ofproto_rule_has_out_port(rule, out_port)
2780 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2781 list_push_back(rules, &rule->ofproto_node);
2787 cls_rule_destroy(&cr);
2791 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2792 * forced into the range of a uint16_t. */
2794 age_secs(long long int age_ms)
2796 return (age_ms < 0 ? 0
2797 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2798 : (unsigned int) age_ms / 1000);
2802 handle_flow_stats_request(struct ofconn *ofconn,
2803 const struct ofp_header *request)
2805 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2806 struct ofputil_flow_stats_request fsr;
2807 struct list replies;
2812 error = ofputil_decode_flow_stats_request(&fsr, request);
2817 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2818 fsr.cookie, fsr.cookie_mask,
2819 fsr.out_port, &rules);
2824 ofpmp_init(&replies, request);
2825 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2826 long long int now = time_msec();
2827 struct ofputil_flow_stats fs;
2829 minimatch_expand(&rule->cr.match, &fs.match);
2830 fs.priority = rule->cr.priority;
2831 fs.cookie = rule->flow_cookie;
2832 fs.table_id = rule->table_id;
2833 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
2834 fs.idle_timeout = rule->idle_timeout;
2835 fs.hard_timeout = rule->hard_timeout;
2836 fs.idle_age = age_secs(now - rule->used);
2837 fs.hard_age = age_secs(now - rule->modified);
2838 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2840 fs.ofpacts = rule->ofpacts;
2841 fs.ofpacts_len = rule->ofpacts_len;
2843 if (rule->send_flow_removed) {
2844 fs.flags |= OFPFF_SEND_FLOW_REM;
2845 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2846 and OFPFF13_NO_BYT_COUNTS */
2848 ofputil_append_flow_stats_reply(&fs, &replies);
2850 ofconn_send_replies(ofconn, &replies);
2856 flow_stats_ds(struct rule *rule, struct ds *results)
2858 uint64_t packet_count, byte_count;
2860 rule->ofproto->ofproto_class->rule_get_stats(rule,
2861 &packet_count, &byte_count);
2863 if (rule->table_id != 0) {
2864 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2866 ds_put_format(results, "duration=%llds, ",
2867 (time_msec() - rule->created) / 1000);
2868 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2869 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2870 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2871 cls_rule_format(&rule->cr, results);
2872 ds_put_char(results, ',');
2873 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2874 ds_put_cstr(results, "\n");
2877 /* Adds a pretty-printed description of all flows to 'results', including
2878 * hidden flows (e.g., set up by in-band control). */
2880 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2882 struct oftable *table;
2884 OFPROTO_FOR_EACH_TABLE (table, p) {
2885 struct cls_cursor cursor;
2888 cls_cursor_init(&cursor, &table->cls, NULL);
2889 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2890 flow_stats_ds(rule, results);
2895 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2896 * '*engine_type' and '*engine_id', respectively. */
2898 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2899 uint8_t *engine_type, uint8_t *engine_id)
2901 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2904 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2905 * true if the port's CFM status was successfully stored into '*status'.
2906 * Returns false if the port did not have CFM configured, in which case
2907 * '*status' is indeterminate.
2909 * The caller must provide and owns '*status', but it does not own and must not
2910 * modify or free the array returned in 'status->rmps'. */
2912 ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
2913 struct ofproto_cfm_status *status)
2915 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2917 && ofproto->ofproto_class->get_cfm_status
2918 && ofproto->ofproto_class->get_cfm_status(ofport, status));
2922 handle_aggregate_stats_request(struct ofconn *ofconn,
2923 const struct ofp_header *oh)
2925 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2926 struct ofputil_flow_stats_request request;
2927 struct ofputil_aggregate_stats stats;
2928 bool unknown_packets, unknown_bytes;
2929 struct ofpbuf *reply;
2934 error = ofputil_decode_flow_stats_request(&request, oh);
2939 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2940 request.cookie, request.cookie_mask,
2941 request.out_port, &rules);
2946 memset(&stats, 0, sizeof stats);
2947 unknown_packets = unknown_bytes = false;
2948 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2949 uint64_t packet_count;
2950 uint64_t byte_count;
2952 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2955 if (packet_count == UINT64_MAX) {
2956 unknown_packets = true;
2958 stats.packet_count += packet_count;
2961 if (byte_count == UINT64_MAX) {
2962 unknown_bytes = true;
2964 stats.byte_count += byte_count;
2969 if (unknown_packets) {
2970 stats.packet_count = UINT64_MAX;
2972 if (unknown_bytes) {
2973 stats.byte_count = UINT64_MAX;
2976 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2977 ofconn_send_reply(ofconn, reply);
2982 struct queue_stats_cbdata {
2983 struct ofport *ofport;
2984 struct list replies;
2988 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2989 const struct netdev_queue_stats *stats)
2992 struct ofputil_queue_stats oqs = {
2993 .port_no = cbdata->ofport->pp.port_no,
2994 .queue_id = queue_id,
2997 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3001 handle_queue_stats_dump_cb(uint32_t queue_id,
3002 struct netdev_queue_stats *stats,
3005 struct queue_stats_cbdata *cbdata = cbdata_;
3007 put_queue_stats(cbdata, queue_id, stats);
3011 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3012 struct queue_stats_cbdata *cbdata)
3014 cbdata->ofport = port;
3015 if (queue_id == OFPQ_ALL) {
3016 netdev_dump_queue_stats(port->netdev,
3017 handle_queue_stats_dump_cb, cbdata);
3019 struct netdev_queue_stats stats;
3021 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3022 put_queue_stats(cbdata, queue_id, &stats);
3024 return OFPERR_OFPQOFC_BAD_QUEUE;
3031 handle_queue_stats_request(struct ofconn *ofconn,
3032 const struct ofp_header *rq)
3034 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3035 struct queue_stats_cbdata cbdata;
3036 struct ofport *port;
3038 struct ofputil_queue_stats_request oqsr;
3040 COVERAGE_INC(ofproto_queue_req);
3042 ofpmp_init(&cbdata.replies, rq);
3044 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3049 if (oqsr.port_no == OFPP_ANY) {
3050 error = OFPERR_OFPQOFC_BAD_QUEUE;
3051 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3052 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3057 port = ofproto_get_port(ofproto, oqsr.port_no);
3059 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3060 : OFPERR_OFPQOFC_BAD_PORT);
3063 ofconn_send_replies(ofconn, &cbdata.replies);
3065 ofpbuf_list_delete(&cbdata.replies);
3072 is_flow_deletion_pending(const struct ofproto *ofproto,
3073 const struct cls_rule *cls_rule,
3076 if (!hmap_is_empty(&ofproto->deletions)) {
3077 struct ofoperation *op;
3079 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3080 cls_rule_hash(cls_rule, table_id),
3081 &ofproto->deletions) {
3082 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3091 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3092 * in which no matching flow already exists in the flow table.
3094 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3095 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3096 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3097 * initiated now but may be retried later.
3099 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3100 * ownership remains with the caller.
3102 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3105 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3106 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3108 struct oftable *table;
3109 struct ofopgroup *group;
3110 struct rule *victim;
3115 error = check_table_id(ofproto, fm->table_id);
3121 if (fm->table_id == 0xff) {
3123 if (ofproto->ofproto_class->rule_choose_table) {
3124 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3130 ovs_assert(table_id < ofproto->n_tables);
3131 table = &ofproto->tables[table_id];
3133 table = &ofproto->tables[0];
3135 } else if (fm->table_id < ofproto->n_tables) {
3136 table = &ofproto->tables[fm->table_id];
3138 return OFPERR_OFPBRC_BAD_TABLE_ID;
3141 if (table->flags & OFTABLE_READONLY) {
3142 return OFPERR_OFPBRC_EPERM;
3145 /* Allocate new rule and initialize classifier rule. */
3146 rule = ofproto->ofproto_class->rule_alloc();
3148 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3149 ofproto->name, strerror(error));
3152 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3154 /* Serialize against pending deletion. */
3155 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3156 cls_rule_destroy(&rule->cr);
3157 ofproto->ofproto_class->rule_dealloc(rule);
3158 return OFPROTO_POSTPONE;
3161 /* Check for overlap, if requested. */
3162 if (fm->flags & OFPFF_CHECK_OVERLAP
3163 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3164 cls_rule_destroy(&rule->cr);
3165 ofproto->ofproto_class->rule_dealloc(rule);
3166 return OFPERR_OFPFMFC_OVERLAP;
3169 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3171 rule->ofproto = ofproto;
3172 rule->pending = NULL;
3173 rule->flow_cookie = fm->new_cookie;
3174 rule->created = rule->modified = rule->used = time_msec();
3175 rule->idle_timeout = fm->idle_timeout;
3176 rule->hard_timeout = fm->hard_timeout;
3177 rule->table_id = table - ofproto->tables;
3178 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3179 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3180 and OFPFF13_NO_BYT_COUNTS */
3181 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3182 rule->ofpacts_len = fm->ofpacts_len;
3183 rule->evictable = true;
3184 rule->eviction_group = NULL;
3185 list_init(&rule->expirable);
3186 rule->monitor_flags = 0;
3187 rule->add_seqno = 0;
3188 rule->modify_seqno = 0;
3190 /* Insert new rule. */
3191 victim = oftable_replace_rule(rule);
3192 if (victim && !rule_is_modifiable(victim)) {
3193 error = OFPERR_OFPBRC_EPERM;
3194 } else if (victim && victim->pending) {
3195 error = OFPROTO_POSTPONE;
3197 struct ofoperation *op;
3200 if (classifier_count(&table->cls) > table->max_flows) {
3203 was_evictable = rule->evictable;
3204 rule->evictable = false;
3205 evict = choose_rule_to_evict(table);
3206 rule->evictable = was_evictable;
3209 error = OFPERR_OFPFMFC_TABLE_FULL;
3211 } else if (evict->pending) {
3212 error = OFPROTO_POSTPONE;
3219 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3220 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3221 op->victim = victim;
3223 error = ofproto->ofproto_class->rule_construct(rule);
3225 op->group->n_running--;
3226 ofoperation_destroy(rule->pending);
3228 delete_flow__(evict, group);
3230 ofopgroup_submit(group);
3234 /* Back out if an error occurred. */
3236 oftable_substitute_rule(rule, victim);
3237 ofproto_rule_destroy__(rule);
3242 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3244 /* Modifies the rules listed in 'rules', changing their actions to match those
3247 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3250 * Returns 0 on success, otherwise an OpenFlow error code. */
3252 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3253 const struct ofputil_flow_mod *fm,
3254 const struct ofp_header *request, struct list *rules)
3256 struct ofopgroup *group;
3260 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3261 error = OFPERR_OFPBRC_EPERM;
3262 LIST_FOR_EACH (rule, ofproto_node, rules) {
3263 struct ofoperation *op;
3264 bool actions_changed;
3265 ovs_be64 new_cookie;
3267 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3269 if (rule_is_modifiable(rule)) {
3270 /* At least one rule is modifiable, don't report EPERM error. */
3276 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3277 rule->ofpacts, rule->ofpacts_len);
3278 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3280 : rule->flow_cookie);
3282 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3283 rule->flow_cookie = new_cookie;
3284 if (actions_changed) {
3285 op->ofpacts = rule->ofpacts;
3286 op->ofpacts_len = rule->ofpacts_len;
3287 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3288 rule->ofpacts_len = fm->ofpacts_len;
3289 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3291 ofoperation_complete(op, 0);
3294 ofopgroup_submit(group);
3300 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3301 const struct ofputil_flow_mod *fm,
3302 const struct ofp_header *request)
3304 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3307 return add_flow(ofproto, ofconn, fm, request);
3310 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3313 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3316 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3317 const struct ofputil_flow_mod *fm,
3318 const struct ofp_header *request)
3323 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3324 fm->cookie, fm->cookie_mask,
3328 } else if (list_is_empty(&rules)) {
3329 return modify_flows_add(ofproto, ofconn, fm, request);
3331 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3335 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3338 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3341 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3342 const struct ofputil_flow_mod *fm,
3343 const struct ofp_header *request)
3348 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3349 fm->priority, fm->cookie, fm->cookie_mask,
3354 } else if (list_is_empty(&rules)) {
3355 return modify_flows_add(ofproto, ofconn, fm, request);
3357 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3358 fm, request, &rules)
3363 /* OFPFC_DELETE implementation. */
3366 delete_flow__(struct rule *rule, struct ofopgroup *group)
3368 struct ofproto *ofproto = rule->ofproto;
3370 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3372 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3373 oftable_remove_rule(rule);
3374 ofproto->ofproto_class->rule_destruct(rule);
3377 /* Deletes the rules listed in 'rules'.
3379 * Returns 0 on success, otherwise an OpenFlow error code. */
3381 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3382 const struct ofp_header *request, struct list *rules)
3384 struct rule *rule, *next;
3385 struct ofopgroup *group;
3387 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3388 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3389 delete_flow__(rule, group);
3391 ofopgroup_submit(group);
3396 /* Implements OFPFC_DELETE. */
3398 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3399 const struct ofputil_flow_mod *fm,
3400 const struct ofp_header *request)
3405 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3406 fm->cookie, fm->cookie_mask,
3407 fm->out_port, &rules);
3408 return (error ? error
3409 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3414 /* Implements OFPFC_DELETE_STRICT. */
3416 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3417 const struct ofputil_flow_mod *fm,
3418 const struct ofp_header *request)
3423 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3424 fm->priority, fm->cookie, fm->cookie_mask,
3425 fm->out_port, &rules);
3426 return (error ? error
3427 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3433 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3435 struct ofputil_flow_removed fr;
3437 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3441 minimatch_expand(&rule->cr.match, &fr.match);
3442 fr.priority = rule->cr.priority;
3443 fr.cookie = rule->flow_cookie;
3445 fr.table_id = rule->table_id;
3446 calc_duration(rule->created, time_msec(),
3447 &fr.duration_sec, &fr.duration_nsec);
3448 fr.idle_timeout = rule->idle_timeout;
3449 fr.hard_timeout = rule->hard_timeout;
3450 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3453 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3457 ofproto_rule_update_used(struct rule *rule, long long int used)
3459 if (used > rule->used) {
3460 struct eviction_group *evg = rule->eviction_group;
3464 heap_change(&evg->rules, &rule->evg_node,
3465 rule_eviction_priority(rule));
3470 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3471 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3474 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3477 * ofproto implementation ->run() functions should use this function to expire
3478 * OpenFlow flows. */
3480 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3482 struct ofproto *ofproto = rule->ofproto;
3483 struct ofopgroup *group;
3485 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3487 ofproto_rule_send_removed(rule, reason);
3489 group = ofopgroup_create_unattached(ofproto);
3490 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3491 oftable_remove_rule(rule);
3492 ofproto->ofproto_class->rule_destruct(rule);
3493 ofopgroup_submit(group);
3497 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3499 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3500 struct ofputil_flow_mod fm;
3501 uint64_t ofpacts_stub[1024 / 8];
3502 struct ofpbuf ofpacts;
3506 error = reject_slave_controller(ofconn);
3511 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3512 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3515 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3516 &fm.match.flow, ofproto->max_ports);
3519 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3522 goto exit_free_ofpacts;
3525 /* Record the operation for logging a summary report. */
3526 switch (fm.command) {
3532 case OFPFC_MODIFY_STRICT:
3533 ofproto->n_modify++;
3537 case OFPFC_DELETE_STRICT:
3538 ofproto->n_delete++;
3543 if (ofproto->next_op_report == LLONG_MAX) {
3544 ofproto->first_op = now;
3545 ofproto->next_op_report = MAX(now + 10 * 1000,
3546 ofproto->op_backoff);
3547 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3549 ofproto->last_op = now;
3552 ofpbuf_uninit(&ofpacts);
3558 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3559 const struct ofputil_flow_mod *fm,
3560 const struct ofp_header *oh)
3562 if (ofproto->n_pending >= 50) {
3563 ovs_assert(!list_is_empty(&ofproto->pending));
3564 return OFPROTO_POSTPONE;
3567 switch (fm->command) {
3569 return add_flow(ofproto, ofconn, fm, oh);
3572 return modify_flows_loose(ofproto, ofconn, fm, oh);
3574 case OFPFC_MODIFY_STRICT:
3575 return modify_flow_strict(ofproto, ofconn, fm, oh);
3578 return delete_flows_loose(ofproto, ofconn, fm, oh);
3580 case OFPFC_DELETE_STRICT:
3581 return delete_flow_strict(ofproto, ofconn, fm, oh);
3584 if (fm->command > 0xff) {
3585 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3586 "flow_mod_table_id extension is not enabled",
3589 return OFPERR_OFPFMFC_BAD_COMMAND;
3594 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3596 struct ofputil_role_request request;
3597 struct ofputil_role_request reply;
3601 error = ofputil_decode_role_message(oh, &request);
3606 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3607 if (ofconn_get_role(ofconn) != request.role
3608 && ofconn_has_pending_opgroups(ofconn)) {
3609 return OFPROTO_POSTPONE;
3612 if (request.have_generation_id
3613 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3614 return OFPERR_OFPRRFC_STALE;
3617 ofconn_set_role(ofconn, request.role);
3620 reply.role = ofconn_get_role(ofconn);
3621 reply.have_generation_id = ofconn_get_master_election_id(
3622 ofconn, &reply.generation_id);
3623 buf = ofputil_encode_role_reply(oh, &reply);
3624 ofconn_send_reply(ofconn, buf);
3630 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3631 const struct ofp_header *oh)
3633 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3634 enum ofputil_protocol cur, next;
3636 cur = ofconn_get_protocol(ofconn);
3637 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3638 ofconn_set_protocol(ofconn, next);
3644 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3646 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3647 enum ofputil_protocol cur, next;
3648 enum ofputil_protocol next_base;
3650 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3652 return OFPERR_OFPBRC_EPERM;
3655 cur = ofconn_get_protocol(ofconn);
3656 next = ofputil_protocol_set_base(cur, next_base);
3657 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3658 /* Avoid sending async messages in surprising protocol. */
3659 return OFPROTO_POSTPONE;
3662 ofconn_set_protocol(ofconn, next);
3667 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3668 const struct ofp_header *oh)
3670 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3673 format = ntohl(msg->format);
3674 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3675 return OFPERR_OFPBRC_EPERM;
3678 if (format != ofconn_get_packet_in_format(ofconn)
3679 && ofconn_has_pending_opgroups(ofconn)) {
3680 /* Avoid sending async message in surprsing packet in format. */
3681 return OFPROTO_POSTPONE;
3684 ofconn_set_packet_in_format(ofconn, format);
3689 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3691 const struct nx_async_config *msg = ofpmsg_body(oh);
3692 uint32_t master[OAM_N_TYPES];
3693 uint32_t slave[OAM_N_TYPES];
3695 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3696 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3697 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3699 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3700 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3701 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3703 ofconn_set_async_config(ofconn, master, slave);
3704 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3705 !ofconn_get_miss_send_len(ofconn)) {
3706 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3713 handle_nxt_set_controller_id(struct ofconn *ofconn,
3714 const struct ofp_header *oh)
3716 const struct nx_controller_id *nci = ofpmsg_body(oh);
3718 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3719 return OFPERR_NXBRC_MUST_BE_ZERO;
3722 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3727 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3731 if (ofconn_has_pending_opgroups(ofconn)) {
3732 return OFPROTO_POSTPONE;
3735 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3736 ? OFPRAW_OFPT10_BARRIER_REPLY
3737 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3738 ofconn_send_reply(ofconn, buf);
3743 ofproto_compose_flow_refresh_update(const struct rule *rule,
3744 enum nx_flow_monitor_flags flags,
3747 struct ofoperation *op = rule->pending;
3748 struct ofputil_flow_update fu;
3751 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3752 /* We'll report the final flow when the operation completes. Reporting
3753 * it now would cause a duplicate report later. */
3757 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3758 ? NXFME_ADDED : NXFME_MODIFIED);
3760 fu.idle_timeout = rule->idle_timeout;
3761 fu.hard_timeout = rule->hard_timeout;
3762 fu.table_id = rule->table_id;
3763 fu.cookie = rule->flow_cookie;
3764 minimatch_expand(&rule->cr.match, &match);
3766 fu.priority = rule->cr.priority;
3767 if (!(flags & NXFMF_ACTIONS)) {
3771 fu.ofpacts = rule->ofpacts;
3772 fu.ofpacts_len = rule->ofpacts_len;
3774 /* An operation is in progress. Use the previous version of the flow's
3775 * actions, so that when the operation commits we report the change. */
3777 case OFOPERATION_ADD:
3778 /* We already verified that there was a victim. */
3779 fu.ofpacts = op->victim->ofpacts;
3780 fu.ofpacts_len = op->victim->ofpacts_len;
3783 case OFOPERATION_MODIFY:
3785 fu.ofpacts = op->ofpacts;
3786 fu.ofpacts_len = op->ofpacts_len;
3788 fu.ofpacts = rule->ofpacts;
3789 fu.ofpacts_len = rule->ofpacts_len;
3793 case OFOPERATION_DELETE:
3794 fu.ofpacts = rule->ofpacts;
3795 fu.ofpacts_len = rule->ofpacts_len;
3803 if (list_is_empty(msgs)) {
3804 ofputil_start_flow_update(msgs);
3806 ofputil_append_flow_update(&fu, msgs);
3810 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3814 LIST_FOR_EACH (rule, ofproto_node, rules) {
3815 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3816 rule->monitor_flags = 0;
3818 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3823 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3824 struct rule *rule, uint64_t seqno,
3827 enum nx_flow_monitor_flags update;
3829 if (ofproto_rule_is_hidden(rule)) {
3834 ? ofoperation_has_out_port(rule->pending, m->out_port)
3835 : ofproto_rule_has_out_port(rule, m->out_port))) {
3840 if (rule->add_seqno > seqno) {
3841 update = NXFMF_ADD | NXFMF_MODIFY;
3842 } else if (rule->modify_seqno > seqno) {
3843 update = NXFMF_MODIFY;
3848 if (!(m->flags & update)) {
3852 update = NXFMF_INITIAL;
3855 if (!rule->monitor_flags) {
3856 list_push_back(rules, &rule->ofproto_node);
3858 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3862 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3866 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3867 const struct ofoperation *op;
3868 const struct oftable *table;
3869 struct cls_rule target;
3871 cls_rule_init_from_minimatch(&target, &m->match, 0);
3872 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3873 struct cls_cursor cursor;
3876 cls_cursor_init(&cursor, &table->cls, &target);
3877 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3878 ovs_assert(!rule->pending); /* XXX */
3879 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3883 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3884 struct rule *rule = op->rule;
3886 if (((m->table_id == 0xff
3887 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3888 : m->table_id == rule->table_id))
3889 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3890 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3893 cls_rule_destroy(&target);
3897 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3900 if (m->flags & NXFMF_INITIAL) {
3901 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3906 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3907 uint64_t seqno, struct list *rules)
3909 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3913 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3915 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3916 struct ofmonitor **monitors;
3917 size_t n_monitors, allocated_monitors;
3918 struct list replies;
3925 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3927 n_monitors = allocated_monitors = 0;
3929 struct ofputil_flow_monitor_request request;
3930 struct ofmonitor *m;
3933 retval = ofputil_decode_flow_monitor_request(&request, &b);
3934 if (retval == EOF) {
3936 } else if (retval) {
3941 if (request.table_id != 0xff
3942 && request.table_id >= ofproto->n_tables) {
3943 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3947 error = ofmonitor_create(&request, ofconn, &m);
3952 if (n_monitors >= allocated_monitors) {
3953 monitors = x2nrealloc(monitors, &allocated_monitors,
3956 monitors[n_monitors++] = m;
3960 for (i = 0; i < n_monitors; i++) {
3961 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3964 ofpmp_init(&replies, oh);
3965 ofmonitor_compose_refresh_updates(&rules, &replies);
3966 ofconn_send_replies(ofconn, &replies);
3973 for (i = 0; i < n_monitors; i++) {
3974 ofmonitor_destroy(monitors[i]);
3981 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3983 struct ofmonitor *m;
3986 id = ofputil_decode_flow_monitor_cancel(oh);
3987 m = ofmonitor_lookup(ofconn, id);
3989 return OFPERR_NXBRC_FM_BAD_ID;
3992 ofmonitor_destroy(m);
3997 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3999 const struct ofp_header *oh = msg->data;
4003 error = ofptype_decode(&type, oh);
4009 /* OpenFlow requests. */
4010 case OFPTYPE_ECHO_REQUEST:
4011 return handle_echo_request(ofconn, oh);
4013 case OFPTYPE_FEATURES_REQUEST:
4014 return handle_features_request(ofconn, oh);
4016 case OFPTYPE_GET_CONFIG_REQUEST:
4017 return handle_get_config_request(ofconn, oh);
4019 case OFPTYPE_SET_CONFIG:
4020 return handle_set_config(ofconn, oh);
4022 case OFPTYPE_PACKET_OUT:
4023 return handle_packet_out(ofconn, oh);
4025 case OFPTYPE_PORT_MOD:
4026 return handle_port_mod(ofconn, oh);
4028 case OFPTYPE_FLOW_MOD:
4029 return handle_flow_mod(ofconn, oh);
4031 case OFPTYPE_BARRIER_REQUEST:
4032 return handle_barrier_request(ofconn, oh);
4034 case OFPTYPE_ROLE_REQUEST:
4035 return handle_role_request(ofconn, oh);
4037 /* OpenFlow replies. */
4038 case OFPTYPE_ECHO_REPLY:
4041 /* Nicira extension requests. */
4042 case OFPTYPE_FLOW_MOD_TABLE_ID:
4043 return handle_nxt_flow_mod_table_id(ofconn, oh);
4045 case OFPTYPE_SET_FLOW_FORMAT:
4046 return handle_nxt_set_flow_format(ofconn, oh);
4048 case OFPTYPE_SET_PACKET_IN_FORMAT:
4049 return handle_nxt_set_packet_in_format(ofconn, oh);
4051 case OFPTYPE_SET_CONTROLLER_ID:
4052 return handle_nxt_set_controller_id(ofconn, oh);
4054 case OFPTYPE_FLOW_AGE:
4055 /* Nothing to do. */
4058 case OFPTYPE_FLOW_MONITOR_CANCEL:
4059 return handle_flow_monitor_cancel(ofconn, oh);
4061 case OFPTYPE_SET_ASYNC_CONFIG:
4062 return handle_nxt_set_async_config(ofconn, oh);
4064 /* Statistics requests. */
4065 case OFPTYPE_DESC_STATS_REQUEST:
4066 return handle_desc_stats_request(ofconn, oh);
4068 case OFPTYPE_FLOW_STATS_REQUEST:
4069 return handle_flow_stats_request(ofconn, oh);
4071 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4072 return handle_aggregate_stats_request(ofconn, oh);
4074 case OFPTYPE_TABLE_STATS_REQUEST:
4075 return handle_table_stats_request(ofconn, oh);
4077 case OFPTYPE_PORT_STATS_REQUEST:
4078 return handle_port_stats_request(ofconn, oh);
4080 case OFPTYPE_QUEUE_STATS_REQUEST:
4081 return handle_queue_stats_request(ofconn, oh);
4083 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4084 return handle_port_desc_stats_request(ofconn, oh);
4086 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4087 return handle_flow_monitor_request(ofconn, oh);
4089 /* FIXME: Change the following once they are implemented: */
4090 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4091 case OFPTYPE_GET_ASYNC_REQUEST:
4092 case OFPTYPE_METER_MOD:
4093 case OFPTYPE_GROUP_REQUEST:
4094 case OFPTYPE_GROUP_DESC_REQUEST:
4095 case OFPTYPE_GROUP_FEATURES_REQUEST:
4096 case OFPTYPE_METER_REQUEST:
4097 case OFPTYPE_METER_CONFIG_REQUEST:
4098 case OFPTYPE_METER_FEATURES_REQUEST:
4099 case OFPTYPE_TABLE_FEATURES_REQUEST:
4100 return OFPERR_OFPBRC_BAD_TYPE;
4104 case OFPTYPE_FEATURES_REPLY:
4105 case OFPTYPE_GET_CONFIG_REPLY:
4106 case OFPTYPE_PACKET_IN:
4107 case OFPTYPE_FLOW_REMOVED:
4108 case OFPTYPE_PORT_STATUS:
4109 case OFPTYPE_BARRIER_REPLY:
4110 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4111 case OFPTYPE_DESC_STATS_REPLY:
4112 case OFPTYPE_FLOW_STATS_REPLY:
4113 case OFPTYPE_QUEUE_STATS_REPLY:
4114 case OFPTYPE_PORT_STATS_REPLY:
4115 case OFPTYPE_TABLE_STATS_REPLY:
4116 case OFPTYPE_AGGREGATE_STATS_REPLY:
4117 case OFPTYPE_PORT_DESC_STATS_REPLY:
4118 case OFPTYPE_ROLE_REPLY:
4119 case OFPTYPE_FLOW_MONITOR_PAUSED:
4120 case OFPTYPE_FLOW_MONITOR_RESUMED:
4121 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4122 case OFPTYPE_GET_ASYNC_REPLY:
4123 case OFPTYPE_GROUP_REPLY:
4124 case OFPTYPE_GROUP_DESC_REPLY:
4125 case OFPTYPE_GROUP_FEATURES_REPLY:
4126 case OFPTYPE_METER_REPLY:
4127 case OFPTYPE_METER_CONFIG_REPLY:
4128 case OFPTYPE_METER_FEATURES_REPLY:
4129 case OFPTYPE_TABLE_FEATURES_REPLY:
4131 return OFPERR_OFPBRC_BAD_TYPE;
4136 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4138 int error = handle_openflow__(ofconn, ofp_msg);
4139 if (error && error != OFPROTO_POSTPONE) {
4140 ofconn_send_error(ofconn, ofp_msg->data, error);
4142 COVERAGE_INC(ofproto_recv_openflow);
4143 return error != OFPROTO_POSTPONE;
4146 /* Asynchronous operations. */
4148 /* Creates and returns a new ofopgroup that is not associated with any
4149 * OpenFlow connection.
4151 * The caller should add operations to the returned group with
4152 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4153 static struct ofopgroup *
4154 ofopgroup_create_unattached(struct ofproto *ofproto)
4156 struct ofopgroup *group = xzalloc(sizeof *group);
4157 group->ofproto = ofproto;
4158 list_init(&group->ofproto_node);
4159 list_init(&group->ops);
4160 list_init(&group->ofconn_node);
4164 /* Creates and returns a new ofopgroup for 'ofproto'.
4166 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4167 * connection. The 'request' and 'buffer_id' arguments are ignored.
4169 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4170 * If the ofopgroup eventually fails, then the error reply will include
4171 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4172 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4174 * The caller should add operations to the returned group with
4175 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4176 static struct ofopgroup *
4177 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4178 const struct ofp_header *request, uint32_t buffer_id)
4180 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4182 size_t request_len = ntohs(request->length);
4184 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4186 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4187 group->ofconn = ofconn;
4188 group->request = xmemdup(request, MIN(request_len, 64));
4189 group->buffer_id = buffer_id;
4194 /* Submits 'group' for processing.
4196 * If 'group' contains no operations (e.g. none were ever added, or all of the
4197 * ones that were added completed synchronously), then it is destroyed
4198 * immediately. Otherwise it is added to the ofproto's list of pending
4201 ofopgroup_submit(struct ofopgroup *group)
4203 if (!group->n_running) {
4204 ofopgroup_complete(group);
4206 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4207 group->ofproto->n_pending++;
4212 ofopgroup_complete(struct ofopgroup *group)
4214 struct ofproto *ofproto = group->ofproto;
4216 struct ofconn *abbrev_ofconn;
4217 ovs_be32 abbrev_xid;
4219 struct ofoperation *op, *next_op;
4222 ovs_assert(!group->n_running);
4225 LIST_FOR_EACH (op, group_node, &group->ops) {
4232 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4233 LIST_FOR_EACH (op, group_node, &group->ops) {
4234 if (op->type != OFOPERATION_DELETE) {
4235 struct ofpbuf *packet;
4238 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4242 error = rule_execute(op->rule, in_port, packet);
4249 if (!error && !list_is_empty(&group->ofconn_node)) {
4250 abbrev_ofconn = group->ofconn;
4251 abbrev_xid = group->request->xid;
4253 abbrev_ofconn = NULL;
4254 abbrev_xid = htonl(0);
4256 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4257 struct rule *rule = op->rule;
4259 /* We generally want to report the change to active OpenFlow flow
4260 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4262 - The operation failed.
4264 - The affected rule is not visible to controllers.
4266 - The operation's only effect was to update rule->modified. */
4268 || ofproto_rule_is_hidden(rule)
4269 || (op->type == OFOPERATION_MODIFY
4271 && rule->flow_cookie == op->flow_cookie))) {
4272 /* Check that we can just cast from ofoperation_type to
4273 * nx_flow_update_event. */
4274 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4276 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4278 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4281 ofmonitor_report(ofproto->connmgr, rule,
4282 (enum nx_flow_update_event) op->type,
4283 op->reason, abbrev_ofconn, abbrev_xid);
4286 rule->pending = NULL;
4289 case OFOPERATION_ADD:
4293 ofproto_rule_destroy__(op->victim);
4294 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4295 if (vid_mask == VLAN_VID_MASK) {
4296 if (ofproto->vlan_bitmap) {
4297 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4298 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4299 bitmap_set1(ofproto->vlan_bitmap, vid);
4300 ofproto->vlans_changed = true;
4303 ofproto->vlans_changed = true;
4307 oftable_substitute_rule(rule, op->victim);
4308 ofproto_rule_destroy__(rule);
4312 case OFOPERATION_DELETE:
4313 ovs_assert(!op->error);
4314 ofproto_rule_destroy__(rule);
4318 case OFOPERATION_MODIFY:
4320 rule->modified = time_msec();
4322 rule->flow_cookie = op->flow_cookie;
4324 free(rule->ofpacts);
4325 rule->ofpacts = op->ofpacts;
4326 rule->ofpacts_len = op->ofpacts_len;
4328 op->ofpacts_len = 0;
4337 ofoperation_destroy(op);
4340 ofmonitor_flush(ofproto->connmgr);
4342 if (!list_is_empty(&group->ofproto_node)) {
4343 ovs_assert(ofproto->n_pending > 0);
4344 ofproto->n_pending--;
4345 list_remove(&group->ofproto_node);
4347 if (!list_is_empty(&group->ofconn_node)) {
4348 list_remove(&group->ofconn_node);
4350 ofconn_send_error(group->ofconn, group->request, error);
4352 connmgr_retry(ofproto->connmgr);
4354 free(group->request);
4358 /* Initiates a new operation on 'rule', of the specified 'type', within
4359 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4361 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4362 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4364 * Returns the newly created ofoperation (which is also available as
4365 * rule->pending). */
4366 static struct ofoperation *
4367 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4368 enum ofoperation_type type,
4369 enum ofp_flow_removed_reason reason)
4371 struct ofproto *ofproto = group->ofproto;
4372 struct ofoperation *op;
4374 ovs_assert(!rule->pending);
4376 op = rule->pending = xzalloc(sizeof *op);
4378 list_push_back(&group->ops, &op->group_node);
4381 op->reason = reason;
4382 op->flow_cookie = rule->flow_cookie;
4386 if (type == OFOPERATION_DELETE) {
4387 hmap_insert(&ofproto->deletions, &op->hmap_node,
4388 cls_rule_hash(&rule->cr, rule->table_id));
4395 ofoperation_destroy(struct ofoperation *op)
4397 struct ofopgroup *group = op->group;
4400 op->rule->pending = NULL;
4402 if (op->type == OFOPERATION_DELETE) {
4403 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4405 list_remove(&op->group_node);
4410 /* Indicates that 'op' completed with status 'error', which is either 0 to
4411 * indicate success or an OpenFlow error code on failure.
4413 * If 'error' is 0, indicating success, the operation will be committed
4414 * permanently to the flow table. There is one interesting subcase:
4416 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4417 * the flow table (the "victim" rule) by a new one, then the caller must
4418 * have uninitialized any derived state in the victim rule, as in step 5 in
4419 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4420 * performs steps 6 and 7 for the victim rule, most notably by calling its
4421 * ->rule_dealloc() function.
4423 * If 'error' is nonzero, then generally the operation will be rolled back:
4425 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4426 * restores the original rule. The caller must have uninitialized any
4427 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4428 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4429 * and 7 for the new rule, calling its ->rule_dealloc() function.
4431 * - If 'op' is a "modify flow" operation, ofproto restores the original
4434 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4435 * allowed to fail. It must always succeed.
4437 * Please see the large comment in ofproto/ofproto-provider.h titled
4438 * "Asynchronous Operation Support" for more information. */
4440 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4442 struct ofopgroup *group = op->group;
4444 ovs_assert(op->rule->pending == op);
4445 ovs_assert(group->n_running > 0);
4446 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4449 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4450 ofopgroup_complete(group);
4455 ofoperation_get_victim(struct ofoperation *op)
4457 ovs_assert(op->type == OFOPERATION_ADD);
4462 pick_datapath_id(const struct ofproto *ofproto)
4464 const struct ofport *port;
4466 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4468 uint8_t ea[ETH_ADDR_LEN];
4471 error = netdev_get_etheraddr(port->netdev, ea);
4473 return eth_addr_to_uint64(ea);
4475 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4476 ofproto->name, netdev_get_name(port->netdev),
4479 return ofproto->fallback_dpid;
4483 pick_fallback_dpid(void)
4485 uint8_t ea[ETH_ADDR_LEN];
4486 eth_addr_nicira_random(ea);
4487 return eth_addr_to_uint64(ea);
4490 /* Table overflow policy. */
4492 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4493 * is not configured to evict rules or if the table contains no evictable
4494 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4496 static struct rule *
4497 choose_rule_to_evict(struct oftable *table)
4499 struct eviction_group *evg;
4501 if (!table->eviction_fields) {
4505 /* In the common case, the outer and inner loops here will each be entered
4508 * - The inner loop normally "return"s in its first iteration. If the
4509 * eviction group has any evictable rules, then it always returns in
4512 * - The outer loop only iterates more than once if the largest eviction
4513 * group has no evictable rules.
4515 * - The outer loop can exit only if table's 'max_flows' is all filled up
4516 * by unevictable rules'. */
4517 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4520 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4521 if (rule->evictable) {
4530 /* Searches 'ofproto' for tables that have more flows than their configured
4531 * maximum and that have flow eviction enabled, and evicts as many flows as
4532 * necessary and currently feasible from them.
4534 * This triggers only when an OpenFlow table has N flows in it and then the
4535 * client configures a maximum number of flows less than N. */
4537 ofproto_evict(struct ofproto *ofproto)
4539 struct ofopgroup *group;
4540 struct oftable *table;
4542 group = ofopgroup_create_unattached(ofproto);
4543 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4544 while (classifier_count(&table->cls) > table->max_flows
4545 && table->eviction_fields) {
4548 rule = choose_rule_to_evict(table);
4549 if (!rule || rule->pending) {
4553 ofoperation_create(group, rule,
4554 OFOPERATION_DELETE, OFPRR_EVICTION);
4555 oftable_remove_rule(rule);
4556 ofproto->ofproto_class->rule_destruct(rule);
4559 ofopgroup_submit(group);
4562 /* Eviction groups. */
4564 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4565 * rules. The priority contains low-order random bits to ensure that eviction
4566 * groups with the same number of rules are prioritized randomly. */
4568 eviction_group_priority(size_t n_rules)
4570 uint16_t size = MIN(UINT16_MAX, n_rules);
4571 return (size << 16) | random_uint16();
4574 /* Updates 'evg', an eviction_group within 'table', following a change that
4575 * adds or removes rules in 'evg'. */
4577 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4579 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4580 eviction_group_priority(heap_count(&evg->rules)));
4583 /* Destroys 'evg', an eviction_group within 'table':
4585 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4586 * rules themselves, just removes them from the eviction group.)
4588 * - Removes 'evg' from 'table'.
4592 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4594 while (!heap_is_empty(&evg->rules)) {
4597 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4598 rule->eviction_group = NULL;
4600 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4601 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4602 heap_destroy(&evg->rules);
4606 /* Removes 'rule' from its eviction group, if any. */
4608 eviction_group_remove_rule(struct rule *rule)
4610 if (rule->eviction_group) {
4611 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4612 struct eviction_group *evg = rule->eviction_group;
4614 rule->eviction_group = NULL;
4615 heap_remove(&evg->rules, &rule->evg_node);
4616 if (heap_is_empty(&evg->rules)) {
4617 eviction_group_destroy(table, evg);
4619 eviction_group_resized(table, evg);
4624 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4625 * returns the hash value. */
4627 eviction_group_hash_rule(struct rule *rule)
4629 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4630 const struct mf_subfield *sf;
4634 hash = table->eviction_group_id_basis;
4635 miniflow_expand(&rule->cr.match.flow, &flow);
4636 for (sf = table->eviction_fields;
4637 sf < &table->eviction_fields[table->n_eviction_fields];
4640 if (mf_are_prereqs_ok(sf->field, &flow)) {
4641 union mf_value value;
4643 mf_get_value(sf->field, &flow, &value);
4645 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4647 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4648 unsigned int start = sf->ofs + sf->n_bits;
4649 bitwise_zero(&value, sf->field->n_bytes, start,
4650 sf->field->n_bytes * 8 - start);
4652 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4654 hash = hash_int(hash, 0);
4661 /* Returns an eviction group within 'table' with the given 'id', creating one
4663 static struct eviction_group *
4664 eviction_group_find(struct oftable *table, uint32_t id)
4666 struct eviction_group *evg;
4668 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4672 evg = xmalloc(sizeof *evg);
4673 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4674 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4675 eviction_group_priority(0));
4676 heap_init(&evg->rules);
4681 /* Returns an eviction priority for 'rule'. The return value should be
4682 * interpreted so that higher priorities make a rule more attractive candidates
4685 rule_eviction_priority(struct rule *rule)
4687 long long int hard_expiration;
4688 long long int idle_expiration;
4689 long long int expiration;
4690 uint32_t expiration_offset;
4692 /* Calculate time of expiration. */
4693 hard_expiration = (rule->hard_timeout
4694 ? rule->modified + rule->hard_timeout * 1000
4696 idle_expiration = (rule->idle_timeout
4697 ? rule->used + rule->idle_timeout * 1000
4699 expiration = MIN(hard_expiration, idle_expiration);
4700 if (expiration == LLONG_MAX) {
4704 /* Calculate the time of expiration as a number of (approximate) seconds
4705 * after program startup.
4707 * This should work OK for program runs that last UINT32_MAX seconds or
4708 * less. Therefore, please restart OVS at least once every 136 years. */
4709 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4711 /* Invert the expiration offset because we're using a max-heap. */
4712 return UINT32_MAX - expiration_offset;
4715 /* Adds 'rule' to an appropriate eviction group for its oftable's
4716 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4717 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4720 * The caller must ensure that 'rule' is not already in an eviction group. */
4722 eviction_group_add_rule(struct rule *rule)
4724 struct ofproto *ofproto = rule->ofproto;
4725 struct oftable *table = &ofproto->tables[rule->table_id];
4727 if (table->eviction_fields
4728 && (rule->hard_timeout || rule->idle_timeout)) {
4729 struct eviction_group *evg;
4731 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4733 rule->eviction_group = evg;
4734 heap_insert(&evg->rules, &rule->evg_node,
4735 rule_eviction_priority(rule));
4736 eviction_group_resized(table, evg);
4742 /* Initializes 'table'. */
4744 oftable_init(struct oftable *table)
4746 memset(table, 0, sizeof *table);
4747 classifier_init(&table->cls);
4748 table->max_flows = UINT_MAX;
4751 /* Destroys 'table', including its classifier and eviction groups.
4753 * The caller is responsible for freeing 'table' itself. */
4755 oftable_destroy(struct oftable *table)
4757 ovs_assert(classifier_is_empty(&table->cls));
4758 oftable_disable_eviction(table);
4759 classifier_destroy(&table->cls);
4763 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4764 * string, then 'table' will use its default name.
4766 * This only affects the name exposed for a table exposed through the OpenFlow
4767 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4769 oftable_set_name(struct oftable *table, const char *name)
4771 if (name && name[0]) {
4772 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4773 if (!table->name || strncmp(name, table->name, len)) {
4775 table->name = xmemdup0(name, len);
4783 /* oftables support a choice of two policies when adding a rule would cause the
4784 * number of flows in the table to exceed the configured maximum number: either
4785 * they can refuse to add the new flow or they can evict some existing flow.
4786 * This function configures the former policy on 'table'. */
4788 oftable_disable_eviction(struct oftable *table)
4790 if (table->eviction_fields) {
4791 struct eviction_group *evg, *next;
4793 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4794 &table->eviction_groups_by_id) {
4795 eviction_group_destroy(table, evg);
4797 hmap_destroy(&table->eviction_groups_by_id);
4798 heap_destroy(&table->eviction_groups_by_size);
4800 free(table->eviction_fields);
4801 table->eviction_fields = NULL;
4802 table->n_eviction_fields = 0;
4806 /* oftables support a choice of two policies when adding a rule would cause the
4807 * number of flows in the table to exceed the configured maximum number: either
4808 * they can refuse to add the new flow or they can evict some existing flow.
4809 * This function configures the latter policy on 'table', with fairness based
4810 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4811 * 'n_fields' as 0 disables fairness.) */
4813 oftable_enable_eviction(struct oftable *table,
4814 const struct mf_subfield *fields, size_t n_fields)
4816 struct cls_cursor cursor;
4819 if (table->eviction_fields
4820 && n_fields == table->n_eviction_fields
4822 || !memcmp(fields, table->eviction_fields,
4823 n_fields * sizeof *fields))) {
4828 oftable_disable_eviction(table);
4830 table->n_eviction_fields = n_fields;
4831 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4833 table->eviction_group_id_basis = random_uint32();
4834 hmap_init(&table->eviction_groups_by_id);
4835 heap_init(&table->eviction_groups_by_size);
4837 cls_cursor_init(&cursor, &table->cls, NULL);
4838 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4839 eviction_group_add_rule(rule);
4843 /* Removes 'rule' from the oftable that contains it. */
4845 oftable_remove_rule(struct rule *rule)
4847 struct ofproto *ofproto = rule->ofproto;
4848 struct oftable *table = &ofproto->tables[rule->table_id];
4850 classifier_remove(&table->cls, &rule->cr);
4851 eviction_group_remove_rule(rule);
4852 if (!list_is_empty(&rule->expirable)) {
4853 list_remove(&rule->expirable);
4857 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4858 * oftable that has an identical cls_rule. Returns the rule that was removed,
4859 * if any, and otherwise NULL. */
4860 static struct rule *
4861 oftable_replace_rule(struct rule *rule)
4863 struct ofproto *ofproto = rule->ofproto;
4864 struct oftable *table = &ofproto->tables[rule->table_id];
4865 struct rule *victim;
4866 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4869 list_insert(&ofproto->expirable, &rule->expirable);
4872 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4874 if (!list_is_empty(&victim->expirable)) {
4875 list_remove(&victim->expirable);
4877 eviction_group_remove_rule(victim);
4879 eviction_group_add_rule(rule);
4883 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4885 oftable_substitute_rule(struct rule *old, struct rule *new)
4888 oftable_replace_rule(new);
4890 oftable_remove_rule(old);
4894 /* unixctl commands. */
4897 ofproto_lookup(const char *name)
4899 struct ofproto *ofproto;
4901 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4903 if (!strcmp(ofproto->name, name)) {
4911 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4912 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4914 struct ofproto *ofproto;
4918 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4919 ds_put_format(&results, "%s\n", ofproto->name);
4921 unixctl_command_reply(conn, ds_cstr(&results));
4922 ds_destroy(&results);
4926 ofproto_unixctl_init(void)
4928 static bool registered;
4934 unixctl_command_register("ofproto/list", "", 0, 0,
4935 ofproto_unixctl_list, NULL);
4938 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4940 * This is deprecated. It is only for compatibility with broken device drivers
4941 * in old versions of Linux that do not properly support VLANs when VLAN
4942 * devices are not used. When broken device drivers are no longer in
4943 * widespread use, we will delete these interfaces. */
4945 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4946 * (exactly) by an OpenFlow rule in 'ofproto'. */
4948 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4950 const struct oftable *oftable;
4952 free(ofproto->vlan_bitmap);
4953 ofproto->vlan_bitmap = bitmap_allocate(4096);
4954 ofproto->vlans_changed = false;
4956 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4957 const struct cls_table *table;
4959 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4960 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4961 const struct cls_rule *rule;
4963 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4964 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4965 bitmap_set1(vlan_bitmap, vid);
4966 bitmap_set1(ofproto->vlan_bitmap, vid);
4973 /* Returns true if new VLANs have come into use by the flow table since the
4974 * last call to ofproto_get_vlan_usage().
4976 * We don't track when old VLANs stop being used. */
4978 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4980 return ofproto->vlans_changed;
4983 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4984 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4985 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4986 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4987 * then the VLAN device is un-enslaved. */
4989 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4990 uint16_t realdev_ofp_port, int vid)
4992 struct ofport *ofport;
4995 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
4997 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4999 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5000 ofproto->name, vlandev_ofp_port);
5004 if (!ofproto->ofproto_class->set_realdev) {
5005 if (!vlandev_ofp_port) {
5008 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5012 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5014 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5015 ofproto->name, vlandev_ofp_port,
5016 netdev_get_name(ofport->netdev), strerror(error));