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 enum ofp_flow_removed_reason);
203 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
204 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
205 const struct ofputil_flow_mod *,
206 const struct ofp_header *);
207 static void calc_duration(long long int start, long long int now,
208 uint32_t *sec, uint32_t *nsec);
211 static uint64_t pick_datapath_id(const struct ofproto *);
212 static uint64_t pick_fallback_dpid(void);
213 static void ofproto_destroy__(struct ofproto *);
214 static void update_mtu(struct ofproto *, struct ofport *);
217 static void ofproto_unixctl_init(void);
219 /* All registered ofproto classes, in probe order. */
220 static const struct ofproto_class **ofproto_classes;
221 static size_t n_ofproto_classes;
222 static size_t allocated_ofproto_classes;
224 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
225 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
227 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
228 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
230 /* Initial mappings of port to OpenFlow number mappings. */
231 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
233 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
235 /* The default value of true waits for flow restore. */
236 static bool flow_restore_wait = true;
238 /* Must be called to initialize the ofproto library.
240 * The caller may pass in 'iface_hints', which contains an shash of
241 * "iface_hint" elements indexed by the interface's name. The provider
242 * may use these hints to describe the startup configuration in order to
243 * reinitialize its state. The caller owns the provided data, so a
244 * provider will make copies of anything required. An ofproto provider
245 * will remove any existing state that is not described by the hint, and
246 * may choose to remove it all. */
248 ofproto_init(const struct shash *iface_hints)
250 struct shash_node *node;
253 ofproto_class_register(&ofproto_dpif_class);
255 /* Make a local copy, since we don't own 'iface_hints' elements. */
256 SHASH_FOR_EACH(node, iface_hints) {
257 const struct iface_hint *orig_hint = node->data;
258 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
259 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
261 new_hint->br_name = xstrdup(orig_hint->br_name);
262 new_hint->br_type = xstrdup(br_type);
263 new_hint->ofp_port = orig_hint->ofp_port;
265 shash_add(&init_ofp_ports, node->name, new_hint);
268 for (i = 0; i < n_ofproto_classes; i++) {
269 ofproto_classes[i]->init(&init_ofp_ports);
273 /* 'type' should be a normalized datapath type, as returned by
274 * ofproto_normalize_type(). Returns the corresponding ofproto_class
275 * structure, or a null pointer if there is none registered for 'type'. */
276 static const struct ofproto_class *
277 ofproto_class_find__(const char *type)
281 for (i = 0; i < n_ofproto_classes; i++) {
282 const struct ofproto_class *class = ofproto_classes[i];
287 class->enumerate_types(&types);
288 found = sset_contains(&types, type);
289 sset_destroy(&types);
295 VLOG_WARN("unknown datapath type %s", type);
299 /* Registers a new ofproto class. After successful registration, new ofprotos
300 * of that type can be created using ofproto_create(). */
302 ofproto_class_register(const struct ofproto_class *new_class)
306 for (i = 0; i < n_ofproto_classes; i++) {
307 if (ofproto_classes[i] == new_class) {
312 if (n_ofproto_classes >= allocated_ofproto_classes) {
313 ofproto_classes = x2nrealloc(ofproto_classes,
314 &allocated_ofproto_classes,
315 sizeof *ofproto_classes);
317 ofproto_classes[n_ofproto_classes++] = new_class;
321 /* Unregisters a datapath provider. 'type' must have been previously
322 * registered and not currently be in use by any ofprotos. After
323 * unregistration new datapaths of that type cannot be opened using
324 * ofproto_create(). */
326 ofproto_class_unregister(const struct ofproto_class *class)
330 for (i = 0; i < n_ofproto_classes; i++) {
331 if (ofproto_classes[i] == class) {
332 for (i++; i < n_ofproto_classes; i++) {
333 ofproto_classes[i - 1] = ofproto_classes[i];
339 VLOG_WARN("attempted to unregister an ofproto class that is not "
344 /* Clears 'types' and enumerates all registered ofproto types into it. The
345 * caller must first initialize the sset. */
347 ofproto_enumerate_types(struct sset *types)
351 for (i = 0; i < n_ofproto_classes; i++) {
352 ofproto_classes[i]->enumerate_types(types);
356 /* Returns the fully spelled out name for the given ofproto 'type'.
358 * Normalized type string can be compared with strcmp(). Unnormalized type
359 * string might be the same even if they have different spellings. */
361 ofproto_normalize_type(const char *type)
363 return type && type[0] ? type : "system";
366 /* Clears 'names' and enumerates the names of all known created ofprotos with
367 * the given 'type'. The caller must first initialize the sset. Returns 0 if
368 * successful, otherwise a positive errno value.
370 * Some kinds of datapaths might not be practically enumerable. This is not
371 * considered an error. */
373 ofproto_enumerate_names(const char *type, struct sset *names)
375 const struct ofproto_class *class = ofproto_class_find__(type);
376 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
380 ofproto_create(const char *datapath_name, const char *datapath_type,
381 struct ofproto **ofprotop)
383 const struct ofproto_class *class;
384 struct ofproto *ofproto;
390 ofproto_unixctl_init();
392 datapath_type = ofproto_normalize_type(datapath_type);
393 class = ofproto_class_find__(datapath_type);
395 VLOG_WARN("could not create datapath %s of unknown type %s",
396 datapath_name, datapath_type);
400 ofproto = class->alloc();
402 VLOG_ERR("failed to allocate datapath %s of type %s",
403 datapath_name, datapath_type);
408 memset(ofproto, 0, sizeof *ofproto);
409 ofproto->ofproto_class = class;
410 ofproto->name = xstrdup(datapath_name);
411 ofproto->type = xstrdup(datapath_type);
412 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
413 hash_string(ofproto->name, 0));
414 ofproto->datapath_id = 0;
415 ofproto->forward_bpdu = false;
416 ofproto->fallback_dpid = pick_fallback_dpid();
417 ofproto->mfr_desc = NULL;
418 ofproto->hw_desc = NULL;
419 ofproto->sw_desc = NULL;
420 ofproto->serial_desc = NULL;
421 ofproto->dp_desc = NULL;
422 ofproto->frag_handling = OFPC_FRAG_NORMAL;
423 hmap_init(&ofproto->ports);
424 shash_init(&ofproto->port_by_name);
425 simap_init(&ofproto->ofp_requests);
426 ofproto->max_ports = OFPP_MAX;
427 ofproto->tables = NULL;
428 ofproto->n_tables = 0;
429 hindex_init(&ofproto->cookies);
430 list_init(&ofproto->expirable);
431 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
432 ofproto->state = S_OPENFLOW;
433 list_init(&ofproto->pending);
434 ofproto->n_pending = 0;
435 hmap_init(&ofproto->deletions);
436 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
437 ofproto->first_op = ofproto->last_op = LLONG_MIN;
438 ofproto->next_op_report = LLONG_MAX;
439 ofproto->op_backoff = LLONG_MIN;
440 ofproto->vlan_bitmap = NULL;
441 ofproto->vlans_changed = false;
442 ofproto->min_mtu = INT_MAX;
444 error = ofproto->ofproto_class->construct(ofproto);
446 VLOG_ERR("failed to open datapath %s: %s",
447 datapath_name, strerror(error));
448 ofproto_destroy__(ofproto);
452 /* The "max_ports" member should have been set by ->construct(ofproto).
453 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
454 ofproto->ofp_port_ids = bitmap_allocate(ofp_to_u16(ofproto->max_ports));
455 bitmap_set1(ofproto->ofp_port_ids, 0);
457 /* Check that hidden tables, if any, are at the end. */
458 ovs_assert(ofproto->n_tables);
459 for (i = 0; i + 1 < ofproto->n_tables; i++) {
460 enum oftable_flags flags = ofproto->tables[i].flags;
461 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
463 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
466 ofproto->datapath_id = pick_datapath_id(ofproto);
473 /* Must be called (only) by an ofproto implementation in its constructor
474 * function. See the large comment on 'construct' in struct ofproto_class for
477 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
479 struct oftable *table;
481 ovs_assert(!ofproto->n_tables);
482 ovs_assert(n_tables >= 1 && n_tables <= 255);
484 ofproto->n_tables = n_tables;
485 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
486 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
491 /* To be optionally called (only) by an ofproto implementation in its
492 * constructor function. See the large comment on 'construct' in struct
493 * ofproto_class for details.
495 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
496 * will then ensure that actions passed into the ofproto implementation will
497 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
498 * function is not called, there will be no such restriction.
500 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
501 * the 'max_ports' restriction. */
503 ofproto_init_max_ports(struct ofproto *ofproto, ofp_port_t max_ports)
505 ovs_assert(ofp_to_u16(max_ports) <= ofp_to_u16(OFPP_MAX));
506 ofproto->max_ports = max_ports;
510 ofproto_get_datapath_id(const struct ofproto *ofproto)
512 return ofproto->datapath_id;
516 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
518 uint64_t old_dpid = p->datapath_id;
519 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
520 if (p->datapath_id != old_dpid) {
521 /* Force all active connections to reconnect, since there is no way to
522 * notify a controller that the datapath ID has changed. */
523 ofproto_reconnect_controllers(p);
528 ofproto_set_controllers(struct ofproto *p,
529 const struct ofproto_controller *controllers,
530 size_t n_controllers, uint32_t allowed_versions)
532 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
537 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
539 connmgr_set_fail_mode(p->connmgr, fail_mode);
542 /* Drops the connections between 'ofproto' and all of its controllers, forcing
543 * them to reconnect. */
545 ofproto_reconnect_controllers(struct ofproto *ofproto)
547 connmgr_reconnect(ofproto->connmgr);
550 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
551 * in-band control should guarantee access, in the same way that in-band
552 * control guarantees access to OpenFlow controllers. */
554 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
555 const struct sockaddr_in *extras, size_t n)
557 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
560 /* Sets the OpenFlow queue used by flows set up by in-band control on
561 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
562 * flows will use the default queue. */
564 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
566 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
569 /* Sets the number of flows at which eviction from the kernel flow table
572 ofproto_set_flow_eviction_threshold(unsigned threshold)
574 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
578 /* Sets the path for handling flow misses. */
580 ofproto_set_flow_miss_model(unsigned model)
582 flow_miss_model = model;
585 /* If forward_bpdu is true, the NORMAL action will forward frames with
586 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
587 * the NORMAL action will drop these frames. */
589 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
591 bool old_val = ofproto->forward_bpdu;
592 ofproto->forward_bpdu = forward_bpdu;
593 if (old_val != ofproto->forward_bpdu) {
594 if (ofproto->ofproto_class->forward_bpdu_changed) {
595 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
600 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
601 * 'idle_time', in seconds, and the maximum number of MAC table entries to
604 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
607 if (ofproto->ofproto_class->set_mac_table_config) {
608 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
614 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
617 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
621 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
623 return connmgr_set_snoops(ofproto->connmgr, snoops);
627 ofproto_set_netflow(struct ofproto *ofproto,
628 const struct netflow_options *nf_options)
630 if (nf_options && sset_is_empty(&nf_options->collectors)) {
634 if (ofproto->ofproto_class->set_netflow) {
635 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
637 return nf_options ? EOPNOTSUPP : 0;
642 ofproto_set_sflow(struct ofproto *ofproto,
643 const struct ofproto_sflow_options *oso)
645 if (oso && sset_is_empty(&oso->targets)) {
649 if (ofproto->ofproto_class->set_sflow) {
650 return ofproto->ofproto_class->set_sflow(ofproto, oso);
652 return oso ? EOPNOTSUPP : 0;
657 ofproto_set_ipfix(struct ofproto *ofproto,
658 const struct ofproto_ipfix_bridge_exporter_options *bo,
659 const struct ofproto_ipfix_flow_exporter_options *fo,
662 if (ofproto->ofproto_class->set_ipfix) {
663 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
665 return (bo || fo) ? EOPNOTSUPP : 0;
670 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
672 flow_restore_wait = flow_restore_wait_db;
676 ofproto_get_flow_restore_wait(void)
678 return flow_restore_wait;
682 /* Spanning Tree Protocol (STP) configuration. */
684 /* Configures STP on 'ofproto' using the settings defined in 's'. If
685 * 's' is NULL, disables STP.
687 * Returns 0 if successful, otherwise a positive errno value. */
689 ofproto_set_stp(struct ofproto *ofproto,
690 const struct ofproto_stp_settings *s)
692 return (ofproto->ofproto_class->set_stp
693 ? ofproto->ofproto_class->set_stp(ofproto, s)
697 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
698 * 'enabled' member of 's' is false, then the other members are not
701 * Returns 0 if successful, otherwise a positive errno value. */
703 ofproto_get_stp_status(struct ofproto *ofproto,
704 struct ofproto_stp_status *s)
706 return (ofproto->ofproto_class->get_stp_status
707 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
711 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
712 * in 's'. The caller is responsible for assigning STP port numbers
713 * (using the 'port_num' member in the range of 1 through 255, inclusive)
714 * and ensuring there are no duplicates. If the 's' is NULL, then STP
715 * is disabled on the port.
717 * Returns 0 if successful, otherwise a positive errno value.*/
719 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
720 const struct ofproto_port_stp_settings *s)
722 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
724 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
725 ofproto->name, ofp_port);
729 return (ofproto->ofproto_class->set_stp_port
730 ? ofproto->ofproto_class->set_stp_port(ofport, s)
734 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
735 * 's'. If the 'enabled' member in 's' is false, then the other members
736 * are not meaningful.
738 * Returns 0 if successful, otherwise a positive errno value.*/
740 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
741 struct ofproto_port_stp_status *s)
743 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
745 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
746 "port %"PRIu16, ofproto->name, ofp_port);
750 return (ofproto->ofproto_class->get_stp_port_status
751 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
755 /* Queue DSCP configuration. */
757 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
758 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
759 * to implement QoS. Instead, it is used to implement features which require
760 * knowledge of what queues exist on a port, and some basic information about
763 * Returns 0 if successful, otherwise a positive errno value. */
765 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
766 const struct ofproto_port_queue *queues,
769 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
772 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
773 ofproto->name, ofp_port);
777 return (ofproto->ofproto_class->set_queues
778 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
782 /* Connectivity Fault Management configuration. */
784 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
786 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
788 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
789 if (ofport && ofproto->ofproto_class->set_cfm) {
790 ofproto->ofproto_class->set_cfm(ofport, NULL);
794 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
795 * basic configuration from the configuration members in 'cfm', and the remote
796 * maintenance point ID from remote_mpid. Ignores the statistics members of
799 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
801 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
802 const struct cfm_settings *s)
804 struct ofport *ofport;
807 ofport = ofproto_get_port(ofproto, ofp_port);
809 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
810 ofproto->name, ofp_port);
814 /* XXX: For configuration simplicity, we only support one remote_mpid
815 * outside of the CFM module. It's not clear if this is the correct long
816 * term solution or not. */
817 error = (ofproto->ofproto_class->set_cfm
818 ? ofproto->ofproto_class->set_cfm(ofport, s)
821 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
822 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
827 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
828 * 'ofproto' does not have a port 'ofp_port'. */
830 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
831 const struct smap *cfg)
833 struct ofport *ofport;
836 ofport = ofproto_get_port(ofproto, ofp_port);
838 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
839 ofproto->name, ofp_port);
843 error = (ofproto->ofproto_class->set_bfd
844 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
847 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
848 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
853 /* Populates 'status' with key value pairs indicating the status of the BFD
854 * session on 'ofp_port'. This information is intended to be populated in the
855 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
858 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
861 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
862 return (ofport && ofproto->ofproto_class->get_bfd_status
863 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
867 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
868 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
869 * 0 if LACP partner information is not current (generally indicating a
870 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
872 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
874 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
875 return (ofport && ofproto->ofproto_class->port_is_lacp_current
876 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
882 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
883 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
884 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
885 * configuration plus, if there is more than one slave, a bonding
888 * If 'aux' is already registered then this function updates its configuration
889 * to 's'. Otherwise, this function registers a new bundle.
891 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
894 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
895 const struct ofproto_bundle_settings *s)
897 return (ofproto->ofproto_class->bundle_set
898 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
902 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
903 * If no such bundle has been registered, this has no effect. */
905 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
907 return ofproto_bundle_register(ofproto, aux, NULL);
911 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
912 * If 'aux' is already registered then this function updates its configuration
913 * to 's'. Otherwise, this function registers a new mirror. */
915 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
916 const struct ofproto_mirror_settings *s)
918 return (ofproto->ofproto_class->mirror_set
919 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
923 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
924 * If no mirror has been registered, this has no effect. */
926 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
928 return ofproto_mirror_register(ofproto, aux, NULL);
931 /* Retrieves statistics from mirror associated with client data pointer
932 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
933 * 'bytes', respectively. If a particular counters is not supported,
934 * the appropriate argument is set to UINT64_MAX. */
936 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
937 uint64_t *packets, uint64_t *bytes)
939 if (!ofproto->ofproto_class->mirror_get_stats) {
940 *packets = *bytes = UINT64_MAX;
944 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
948 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
949 * which all packets are flooded, instead of using MAC learning. If
950 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
952 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
955 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
957 return (ofproto->ofproto_class->set_flood_vlans
958 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
962 /* Returns true if 'aux' is a registered bundle that is currently in use as the
963 * output for a mirror. */
965 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
967 return (ofproto->ofproto_class->is_mirror_output_bundle
968 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
972 /* Configuration of OpenFlow tables. */
974 /* Returns the number of OpenFlow tables in 'ofproto'. */
976 ofproto_get_n_tables(const struct ofproto *ofproto)
978 return ofproto->n_tables;
981 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
982 * settings from 's'. 'table_id' must be in the range 0 through the number of
983 * OpenFlow tables in 'ofproto' minus 1, inclusive.
985 * For read-only tables, only the name may be configured. */
987 ofproto_configure_table(struct ofproto *ofproto, int table_id,
988 const struct ofproto_table_settings *s)
990 struct oftable *table;
992 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
993 table = &ofproto->tables[table_id];
995 oftable_set_name(table, s->name);
997 if (table->flags & OFTABLE_READONLY) {
1002 oftable_enable_eviction(table, s->groups, s->n_groups);
1004 oftable_disable_eviction(table);
1007 table->max_flows = s->max_flows;
1008 if (classifier_count(&table->cls) > table->max_flows
1009 && table->eviction_fields) {
1010 /* 'table' contains more flows than allowed. We might not be able to
1011 * evict them right away because of the asynchronous nature of flow
1012 * table changes. Schedule eviction for later. */
1013 switch (ofproto->state) {
1015 ofproto->state = S_EVICT;
1019 /* We're already deleting flows, nothing more to do. */
1026 ofproto_has_snoops(const struct ofproto *ofproto)
1028 return connmgr_has_snoops(ofproto->connmgr);
1032 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1034 connmgr_get_snoops(ofproto->connmgr, snoops);
1038 ofproto_flush__(struct ofproto *ofproto)
1040 struct ofopgroup *group;
1041 struct oftable *table;
1043 if (ofproto->ofproto_class->flush) {
1044 ofproto->ofproto_class->flush(ofproto);
1047 group = ofopgroup_create_unattached(ofproto);
1048 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1049 struct rule *rule, *next_rule;
1050 struct cls_cursor cursor;
1052 if (table->flags & OFTABLE_HIDDEN) {
1056 cls_cursor_init(&cursor, &table->cls, NULL);
1057 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1058 if (!rule->pending) {
1059 ofoperation_create(group, rule, OFOPERATION_DELETE,
1061 oftable_remove_rule(rule);
1062 ofproto->ofproto_class->rule_destruct(rule);
1066 ofopgroup_submit(group);
1070 ofproto_destroy__(struct ofproto *ofproto)
1072 struct oftable *table;
1074 ovs_assert(list_is_empty(&ofproto->pending));
1075 ovs_assert(!ofproto->n_pending);
1077 connmgr_destroy(ofproto->connmgr);
1079 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1080 free(ofproto->name);
1081 free(ofproto->type);
1082 free(ofproto->mfr_desc);
1083 free(ofproto->hw_desc);
1084 free(ofproto->sw_desc);
1085 free(ofproto->serial_desc);
1086 free(ofproto->dp_desc);
1087 hmap_destroy(&ofproto->ports);
1088 shash_destroy(&ofproto->port_by_name);
1089 bitmap_free(ofproto->ofp_port_ids);
1090 simap_destroy(&ofproto->ofp_requests);
1092 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1093 oftable_destroy(table);
1095 free(ofproto->tables);
1097 hmap_destroy(&ofproto->deletions);
1099 free(ofproto->vlan_bitmap);
1101 ofproto->ofproto_class->dealloc(ofproto);
1105 ofproto_destroy(struct ofproto *p)
1107 struct ofport *ofport, *next_ofport;
1114 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1115 ofport_destroy(ofport);
1118 p->ofproto_class->destruct(p);
1119 ofproto_destroy__(p);
1122 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1123 * kernel datapath, for example, this destroys the datapath in the kernel, and
1124 * with the netdev-based datapath, it tears down the data structures that
1125 * represent the datapath.
1127 * The datapath should not be currently open as an ofproto. */
1129 ofproto_delete(const char *name, const char *type)
1131 const struct ofproto_class *class = ofproto_class_find__(type);
1132 return (!class ? EAFNOSUPPORT
1133 : !class->del ? EACCES
1134 : class->del(type, name));
1138 process_port_change(struct ofproto *ofproto, int error, char *devname)
1140 if (error == ENOBUFS) {
1141 reinit_ports(ofproto);
1142 } else if (!error) {
1143 update_port(ofproto, devname);
1149 ofproto_type_run(const char *datapath_type)
1151 const struct ofproto_class *class;
1154 datapath_type = ofproto_normalize_type(datapath_type);
1155 class = ofproto_class_find__(datapath_type);
1157 error = class->type_run ? class->type_run(datapath_type) : 0;
1158 if (error && error != EAGAIN) {
1159 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1160 datapath_type, strerror(error));
1166 ofproto_type_run_fast(const char *datapath_type)
1168 const struct ofproto_class *class;
1171 datapath_type = ofproto_normalize_type(datapath_type);
1172 class = ofproto_class_find__(datapath_type);
1174 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1175 if (error && error != EAGAIN) {
1176 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1177 datapath_type, strerror(error));
1183 ofproto_type_wait(const char *datapath_type)
1185 const struct ofproto_class *class;
1187 datapath_type = ofproto_normalize_type(datapath_type);
1188 class = ofproto_class_find__(datapath_type);
1190 if (class->type_wait) {
1191 class->type_wait(datapath_type);
1196 ofproto_run(struct ofproto *p)
1198 struct sset changed_netdevs;
1199 const char *changed_netdev;
1200 struct ofport *ofport;
1203 error = p->ofproto_class->run(p);
1204 if (error && error != EAGAIN) {
1205 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1208 if (p->ofproto_class->port_poll) {
1211 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1212 process_port_change(p, error, devname);
1216 /* Update OpenFlow port status for any port whose netdev has changed.
1218 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1219 * destroyed, so it's not safe to update ports directly from the
1220 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1221 * need this two-phase approach. */
1222 sset_init(&changed_netdevs);
1223 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1224 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1225 if (ofport->change_seq != change_seq) {
1226 ofport->change_seq = change_seq;
1227 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1230 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1231 update_port(p, changed_netdev);
1233 sset_destroy(&changed_netdevs);
1237 connmgr_run(p->connmgr, handle_openflow);
1241 connmgr_run(p->connmgr, NULL);
1243 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1244 p->state = S_OPENFLOW;
1249 connmgr_run(p->connmgr, NULL);
1251 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1252 connmgr_flushed(p->connmgr);
1253 p->state = S_OPENFLOW;
1261 if (time_msec() >= p->next_op_report) {
1262 long long int ago = (time_msec() - p->first_op) / 1000;
1263 long long int interval = (p->last_op - p->first_op) / 1000;
1267 ds_put_format(&s, "%d flow_mods ",
1268 p->n_add + p->n_delete + p->n_modify);
1269 if (interval == ago) {
1270 ds_put_format(&s, "in the last %lld s", ago);
1271 } else if (interval) {
1272 ds_put_format(&s, "in the %lld s starting %lld s ago",
1275 ds_put_format(&s, "%lld s ago", ago);
1278 ds_put_cstr(&s, " (");
1280 ds_put_format(&s, "%d adds, ", p->n_add);
1283 ds_put_format(&s, "%d deletes, ", p->n_delete);
1286 ds_put_format(&s, "%d modifications, ", p->n_modify);
1289 ds_put_char(&s, ')');
1291 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1294 p->n_add = p->n_delete = p->n_modify = 0;
1295 p->next_op_report = LLONG_MAX;
1301 /* Performs periodic activity required by 'ofproto' that needs to be done
1302 * with the least possible latency.
1304 * It makes sense to call this function a couple of times per poll loop, to
1305 * provide a significant performance boost on some benchmarks with the
1306 * ofproto-dpif implementation. */
1308 ofproto_run_fast(struct ofproto *p)
1312 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1313 if (error && error != EAGAIN) {
1314 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1315 p->name, strerror(error));
1321 ofproto_wait(struct ofproto *p)
1323 struct ofport *ofport;
1325 p->ofproto_class->wait(p);
1326 if (p->ofproto_class->port_poll_wait) {
1327 p->ofproto_class->port_poll_wait(p);
1330 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1331 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1332 poll_immediate_wake();
1338 connmgr_wait(p->connmgr, true);
1343 connmgr_wait(p->connmgr, false);
1344 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1345 poll_immediate_wake();
1352 ofproto_is_alive(const struct ofproto *p)
1354 return connmgr_has_controllers(p->connmgr);
1357 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1358 * memory_report(). */
1360 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1362 const struct oftable *table;
1363 unsigned int n_rules;
1365 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1366 simap_increase(usage, "ops",
1367 ofproto->n_pending + hmap_count(&ofproto->deletions));
1370 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1371 n_rules += classifier_count(&table->cls);
1373 simap_increase(usage, "rules", n_rules);
1375 if (ofproto->ofproto_class->get_memory_usage) {
1376 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1379 connmgr_get_memory_usage(ofproto->connmgr, usage);
1383 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1386 connmgr_get_controller_info(ofproto->connmgr, info);
1390 ofproto_free_ofproto_controller_info(struct shash *info)
1392 connmgr_free_controller_info(info);
1395 /* Makes a deep copy of 'old' into 'port'. */
1397 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1399 port->name = xstrdup(old->name);
1400 port->type = xstrdup(old->type);
1401 port->ofp_port = old->ofp_port;
1404 /* Frees memory allocated to members of 'ofproto_port'.
1406 * Do not call this function on an ofproto_port obtained from
1407 * ofproto_port_dump_next(): that function retains ownership of the data in the
1410 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1412 free(ofproto_port->name);
1413 free(ofproto_port->type);
1416 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1418 * This function provides no status indication. An error status for the entire
1419 * dump operation is provided when it is completed by calling
1420 * ofproto_port_dump_done().
1423 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1424 const struct ofproto *ofproto)
1426 dump->ofproto = ofproto;
1427 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1431 /* Attempts to retrieve another port from 'dump', which must have been created
1432 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1433 * 'port' and returns true. On failure, returns false.
1435 * Failure might indicate an actual error or merely that the last port has been
1436 * dumped. An error status for the entire dump operation is provided when it
1437 * is completed by calling ofproto_port_dump_done().
1439 * The ofproto owns the data stored in 'port'. It will remain valid until at
1440 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1441 * ofproto_port_dump_done(). */
1443 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1444 struct ofproto_port *port)
1446 const struct ofproto *ofproto = dump->ofproto;
1452 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1455 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1461 /* Completes port table dump operation 'dump', which must have been created
1462 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1463 * error-free, otherwise a positive errno value describing the problem. */
1465 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1467 const struct ofproto *ofproto = dump->ofproto;
1469 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1472 return dump->error == EOF ? 0 : dump->error;
1475 /* Returns the type to pass to netdev_open() when a datapath of type
1476 * 'datapath_type' has a port of type 'port_type', for a few special
1477 * cases when a netdev type differs from a port type. For example, when
1478 * using the userspace datapath, a port of type "internal" needs to be
1481 * Returns either 'type' itself or a string literal, which must not be
1484 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1486 const struct ofproto_class *class;
1488 datapath_type = ofproto_normalize_type(datapath_type);
1489 class = ofproto_class_find__(datapath_type);
1494 return (class->port_open_type
1495 ? class->port_open_type(datapath_type, port_type)
1499 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1500 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1501 * the port's OpenFlow port number.
1503 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1504 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1505 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1506 * 'ofp_portp' is non-null). */
1508 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1509 ofp_port_t *ofp_portp)
1511 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1514 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1516 const char *netdev_name = netdev_get_name(netdev);
1518 simap_put(&ofproto->ofp_requests, netdev_name,
1519 ofp_to_u16(ofp_port));
1520 update_port(ofproto, netdev_name);
1523 struct ofproto_port ofproto_port;
1525 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1527 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1528 ofproto_port_destroy(&ofproto_port);
1533 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1534 * initializes '*port' appropriately; on failure, returns a positive errno
1537 * The caller owns the data in 'ofproto_port' and must free it with
1538 * ofproto_port_destroy() when it is no longer needed. */
1540 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1541 struct ofproto_port *port)
1545 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1547 memset(port, 0, sizeof *port);
1552 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1553 * Returns 0 if successful, otherwise a positive errno. */
1555 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1557 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1558 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1559 struct simap_node *ofp_request_node;
1562 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1563 if (ofp_request_node) {
1564 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1567 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1568 if (!error && ofport) {
1569 /* 'name' is the netdev's name and update_port() is going to close the
1570 * netdev. Just in case update_port() refers to 'name' after it
1571 * destroys 'ofport', make a copy of it around the update_port()
1573 char *devname = xstrdup(name);
1574 update_port(ofproto, devname);
1580 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1581 * performs the 'n_actions' actions in 'actions'. The new flow will not
1584 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1585 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1586 * controllers; otherwise, it will be hidden.
1588 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1590 * This is a helper function for in-band control and fail-open. */
1592 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1593 unsigned int priority,
1594 const struct ofpact *ofpacts, size_t ofpacts_len)
1596 const struct rule *rule;
1598 rule = rule_from_cls_rule(classifier_find_match_exactly(
1599 &ofproto->tables[0].cls, match, priority));
1600 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1601 ofpacts, ofpacts_len)) {
1602 struct ofputil_flow_mod fm;
1604 memset(&fm, 0, sizeof fm);
1606 fm.priority = priority;
1607 fm.buffer_id = UINT32_MAX;
1608 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1609 fm.ofpacts_len = ofpacts_len;
1610 add_flow(ofproto, NULL, &fm, NULL);
1615 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1616 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1617 * operation cannot be initiated now but may be retried later.
1619 * This is a helper function for in-band control and fail-open. */
1621 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1623 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1626 /* Searches for a rule with matching criteria exactly equal to 'target' in
1627 * ofproto's table 0 and, if it finds one, deletes it.
1629 * This is a helper function for in-band control and fail-open. */
1631 ofproto_delete_flow(struct ofproto *ofproto,
1632 const struct match *target, unsigned int priority)
1636 rule = rule_from_cls_rule(classifier_find_match_exactly(
1637 &ofproto->tables[0].cls, target, priority));
1639 /* No such rule -> success. */
1641 } else if (rule->pending) {
1642 /* An operation on the rule is already pending -> failure.
1643 * Caller must retry later if it's important. */
1646 /* Initiate deletion -> success. */
1647 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1648 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1649 oftable_remove_rule(rule);
1650 ofproto->ofproto_class->rule_destruct(rule);
1651 ofopgroup_submit(group);
1657 /* Starts the process of deleting all of the flows from all of ofproto's flow
1658 * tables and then reintroducing the flows required by in-band control and
1659 * fail-open. The process will complete in a later call to ofproto_run(). */
1661 ofproto_flush_flows(struct ofproto *ofproto)
1663 COVERAGE_INC(ofproto_flush);
1664 ofproto->state = S_FLUSH;
1668 reinit_ports(struct ofproto *p)
1670 struct ofproto_port_dump dump;
1671 struct sset devnames;
1672 struct ofport *ofport;
1673 struct ofproto_port ofproto_port;
1674 const char *devname;
1676 COVERAGE_INC(ofproto_reinit_ports);
1678 sset_init(&devnames);
1679 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1680 sset_add(&devnames, netdev_get_name(ofport->netdev));
1682 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1683 sset_add(&devnames, ofproto_port.name);
1686 SSET_FOR_EACH (devname, &devnames) {
1687 update_port(p, devname);
1689 sset_destroy(&devnames);
1693 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1695 uint16_t max_ports = ofp_to_u16(ofproto->max_ports);
1698 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1700 port_idx = UINT16_MAX;
1703 if (port_idx >= max_ports
1704 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1705 uint16_t end_port_no = ofp_to_u16(ofproto->alloc_port_no);
1706 uint16_t alloc_port_no = end_port_no;
1708 /* Search for a free OpenFlow port number. We try not to
1709 * immediately reuse them to prevent problems due to old
1712 if (++alloc_port_no >= max_ports) {
1715 if (!bitmap_is_set(ofproto->ofp_port_ids, alloc_port_no)) {
1716 port_idx = alloc_port_no;
1717 ofproto->alloc_port_no = u16_to_ofp(alloc_port_no);
1720 if (alloc_port_no == end_port_no) {
1725 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1726 return u16_to_ofp(port_idx);
1730 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1732 if (ofp_to_u16(ofp_port) < ofp_to_u16(ofproto->max_ports)) {
1733 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1737 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1738 * pointer if the netdev cannot be opened. On success, also fills in
1740 static struct netdev *
1741 ofport_open(struct ofproto *ofproto,
1742 struct ofproto_port *ofproto_port,
1743 struct ofputil_phy_port *pp)
1745 enum netdev_flags flags;
1746 struct netdev *netdev;
1749 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1751 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1752 "cannot be opened (%s)",
1754 ofproto_port->name, ofproto_port->ofp_port,
1755 ofproto_port->name, strerror(error));
1759 if (ofproto_port->ofp_port == OFPP_NONE) {
1760 if (!strcmp(ofproto->name, ofproto_port->name)) {
1761 ofproto_port->ofp_port = OFPP_LOCAL;
1763 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1764 ofproto_port->name);
1767 pp->port_no = ofproto_port->ofp_port;
1768 netdev_get_etheraddr(netdev, pp->hw_addr);
1769 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1770 netdev_get_flags(netdev, &flags);
1771 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1772 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1773 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1774 &pp->supported, &pp->peer);
1775 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1776 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1781 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1782 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1785 ofport_equal(const struct ofputil_phy_port *a,
1786 const struct ofputil_phy_port *b)
1788 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1789 && a->state == b->state
1790 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1791 && a->curr == b->curr
1792 && a->advertised == b->advertised
1793 && a->supported == b->supported
1794 && a->peer == b->peer
1795 && a->curr_speed == b->curr_speed
1796 && a->max_speed == b->max_speed);
1799 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1800 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1801 * one with the same name or port number). */
1803 ofport_install(struct ofproto *p,
1804 struct netdev *netdev, const struct ofputil_phy_port *pp)
1806 const char *netdev_name = netdev_get_name(netdev);
1807 struct ofport *ofport;
1810 /* Create ofport. */
1811 ofport = p->ofproto_class->port_alloc();
1816 ofport->ofproto = p;
1817 ofport->netdev = netdev;
1818 ofport->change_seq = netdev_change_seq(netdev);
1820 ofport->ofp_port = pp->port_no;
1821 ofport->created = time_msec();
1823 /* Add port to 'p'. */
1824 hmap_insert(&p->ports, &ofport->hmap_node,
1825 hash_ofp_port(ofport->ofp_port));
1826 shash_add(&p->port_by_name, netdev_name, ofport);
1828 update_mtu(p, ofport);
1830 /* Let the ofproto_class initialize its private data. */
1831 error = p->ofproto_class->port_construct(ofport);
1835 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1839 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1840 p->name, netdev_name, strerror(error));
1842 ofport_destroy__(ofport);
1844 netdev_close(netdev);
1848 /* Removes 'ofport' from 'p' and destroys it. */
1850 ofport_remove(struct ofport *ofport)
1852 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1854 ofport_destroy(ofport);
1857 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1860 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1862 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1864 ofport_remove(port);
1868 /* Updates 'port' with new 'pp' description.
1870 * Does not handle a name or port number change. The caller must implement
1871 * such a change as a delete followed by an add. */
1873 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1875 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1876 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1877 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1878 port->pp.state = pp->state;
1879 port->pp.curr = pp->curr;
1880 port->pp.advertised = pp->advertised;
1881 port->pp.supported = pp->supported;
1882 port->pp.peer = pp->peer;
1883 port->pp.curr_speed = pp->curr_speed;
1884 port->pp.max_speed = pp->max_speed;
1886 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1889 /* Update OpenFlow 'state' in 'port' and notify controller. */
1891 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1893 if (port->pp.state != state) {
1894 port->pp.state = state;
1895 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1901 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
1903 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1905 if (port->ofproto->ofproto_class->set_realdev) {
1906 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1908 if (port->ofproto->ofproto_class->set_stp_port) {
1909 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1911 if (port->ofproto->ofproto_class->set_cfm) {
1912 port->ofproto->ofproto_class->set_cfm(port, NULL);
1914 if (port->ofproto->ofproto_class->bundle_remove) {
1915 port->ofproto->ofproto_class->bundle_remove(port);
1921 ofport_destroy__(struct ofport *port)
1923 struct ofproto *ofproto = port->ofproto;
1924 const char *name = netdev_get_name(port->netdev);
1926 hmap_remove(&ofproto->ports, &port->hmap_node);
1927 shash_delete(&ofproto->port_by_name,
1928 shash_find(&ofproto->port_by_name, name));
1930 netdev_close(port->netdev);
1931 ofproto->ofproto_class->port_dealloc(port);
1935 ofport_destroy(struct ofport *port)
1938 dealloc_ofp_port(port->ofproto, port->ofp_port);
1939 port->ofproto->ofproto_class->port_destruct(port);
1940 ofport_destroy__(port);
1945 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1947 struct ofport *port;
1949 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
1951 if (port->ofp_port == ofp_port) {
1959 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1961 struct ofproto *ofproto = port->ofproto;
1964 if (ofproto->ofproto_class->port_get_stats) {
1965 error = ofproto->ofproto_class->port_get_stats(port, stats);
1974 update_port(struct ofproto *ofproto, const char *name)
1976 struct ofproto_port ofproto_port;
1977 struct ofputil_phy_port pp;
1978 struct netdev *netdev;
1979 struct ofport *port;
1981 COVERAGE_INC(ofproto_update_port);
1983 /* Fetch 'name''s location and properties from the datapath. */
1984 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1985 ? ofport_open(ofproto, &ofproto_port, &pp)
1989 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1990 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1991 struct netdev *old_netdev = port->netdev;
1993 /* 'name' hasn't changed location. Any properties changed? */
1994 if (!ofport_equal(&port->pp, &pp)) {
1995 ofport_modified(port, &pp);
1998 update_mtu(ofproto, port);
2000 /* Install the newly opened netdev in case it has changed.
2001 * Don't close the old netdev yet in case port_modified has to
2002 * remove a retained reference to it.*/
2003 port->netdev = netdev;
2004 port->change_seq = netdev_change_seq(netdev);
2006 if (port->ofproto->ofproto_class->port_modified) {
2007 port->ofproto->ofproto_class->port_modified(port);
2010 netdev_close(old_netdev);
2012 /* If 'port' is nonnull then its name differs from 'name' and thus
2013 * we should delete it. If we think there's a port named 'name'
2014 * then its port number must be wrong now so delete it too. */
2016 ofport_remove(port);
2018 ofport_remove_with_name(ofproto, name);
2019 ofport_install(ofproto, netdev, &pp);
2022 /* Any port named 'name' is gone now. */
2023 ofport_remove_with_name(ofproto, name);
2025 ofproto_port_destroy(&ofproto_port);
2029 init_ports(struct ofproto *p)
2031 struct ofproto_port_dump dump;
2032 struct ofproto_port ofproto_port;
2033 struct shash_node *node, *next;
2035 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2036 const char *name = ofproto_port.name;
2038 if (shash_find(&p->port_by_name, name)) {
2039 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2042 struct ofputil_phy_port pp;
2043 struct netdev *netdev;
2045 /* Check if an OpenFlow port number had been requested. */
2046 node = shash_find(&init_ofp_ports, name);
2048 const struct iface_hint *iface_hint = node->data;
2049 simap_put(&p->ofp_requests, name,
2050 ofp_to_u16(iface_hint->ofp_port));
2053 netdev = ofport_open(p, &ofproto_port, &pp);
2055 ofport_install(p, netdev, &pp);
2060 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2061 struct iface_hint *iface_hint = node->data;
2063 if (!strcmp(iface_hint->br_name, p->name)) {
2064 free(iface_hint->br_name);
2065 free(iface_hint->br_type);
2067 shash_delete(&init_ofp_ports, node);
2074 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2075 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2077 find_min_mtu(struct ofproto *p)
2079 struct ofport *ofport;
2082 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2083 struct netdev *netdev = ofport->netdev;
2086 /* Skip any internal ports, since that's what we're trying to
2088 if (!strcmp(netdev_get_type(netdev), "internal")) {
2092 if (netdev_get_mtu(netdev, &dev_mtu)) {
2095 if (!mtu || dev_mtu < mtu) {
2100 return mtu ? mtu: ETH_PAYLOAD_MAX;
2103 /* Update MTU of all datapath devices on 'p' to the minimum of the
2104 * non-datapath ports in event of 'port' added or changed. */
2106 update_mtu(struct ofproto *p, struct ofport *port)
2108 struct ofport *ofport;
2109 struct netdev *netdev = port->netdev;
2110 int dev_mtu, old_min;
2112 if (netdev_get_mtu(netdev, &dev_mtu)) {
2116 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2117 if (dev_mtu > p->min_mtu) {
2118 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2119 dev_mtu = p->min_mtu;
2122 port->mtu = dev_mtu;
2126 /* For non-internal port find new min mtu. */
2127 old_min = p->min_mtu;
2128 port->mtu = dev_mtu;
2129 p->min_mtu = find_min_mtu(p);
2130 if (p->min_mtu == old_min) {
2134 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2135 struct netdev *netdev = ofport->netdev;
2137 if (!strcmp(netdev_get_type(netdev), "internal")) {
2138 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2139 ofport->mtu = p->min_mtu;
2146 ofproto_rule_destroy__(struct rule *rule)
2149 cls_rule_destroy(&rule->cr);
2150 free(rule->ofpacts);
2151 rule->ofproto->ofproto_class->rule_dealloc(rule);
2155 /* This function allows an ofproto implementation to destroy any rules that
2156 * remain when its ->destruct() function is called. The caller must have
2157 * already uninitialized any derived members of 'rule' (step 5 described in the
2158 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2159 * This function implements steps 6 and 7.
2161 * This function should only be called from an ofproto implementation's
2162 * ->destruct() function. It is not suitable elsewhere. */
2164 ofproto_rule_destroy(struct rule *rule)
2166 ovs_assert(!rule->pending);
2167 oftable_remove_rule(rule);
2168 ofproto_rule_destroy__(rule);
2171 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2172 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2174 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2176 return (port == OFPP_ANY
2177 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2180 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2181 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2183 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2185 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2190 case OFOPERATION_ADD:
2191 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2193 case OFOPERATION_DELETE:
2196 case OFOPERATION_MODIFY:
2197 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2203 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2204 * statistics appropriately. 'packet' must have at least sizeof(struct
2205 * ofp10_packet_in) bytes of headroom.
2207 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2208 * with statistics for 'packet' either way.
2210 * Takes ownership of 'packet'. */
2212 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2215 union flow_in_port in_port_;
2217 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2219 in_port_.ofp_port = in_port;
2220 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2221 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2224 /* Returns true if 'rule' should be hidden from the controller.
2226 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2227 * (e.g. by in-band control) and are intentionally hidden from the
2230 ofproto_rule_is_hidden(const struct rule *rule)
2232 return rule->cr.priority > UINT16_MAX;
2235 static enum oftable_flags
2236 rule_get_flags(const struct rule *rule)
2238 return rule->ofproto->tables[rule->table_id].flags;
2242 rule_is_modifiable(const struct rule *rule)
2244 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2248 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2250 ofconn_send_reply(ofconn, make_echo_reply(oh));
2255 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2257 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2258 struct ofputil_switch_features features;
2259 struct ofport *port;
2265 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2267 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2269 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2270 * if present, are always at the end.) */
2271 n_tables = ofproto->n_tables;
2272 for (i = 0; i < ofproto->n_tables; i++) {
2273 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2279 features.datapath_id = ofproto->datapath_id;
2280 features.n_buffers = pktbuf_capacity();
2281 features.n_tables = n_tables;
2282 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2283 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2285 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2287 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2288 features.auxiliary_id = 0;
2289 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2291 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2292 ofputil_put_switch_features_port(&port->pp, b);
2295 ofconn_send_reply(ofconn, b);
2300 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2302 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2303 struct ofp_switch_config *osc;
2304 enum ofp_config_flags flags;
2308 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2309 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2310 flags = ofproto->frag_handling;
2311 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2312 if (oh->version < OFP13_VERSION
2313 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2314 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2316 osc->flags = htons(flags);
2317 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2318 ofconn_send_reply(ofconn, buf);
2324 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2326 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2327 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2328 uint16_t flags = ntohs(osc->flags);
2330 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2331 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2332 enum ofp_config_flags cur = ofproto->frag_handling;
2333 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2335 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2337 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2338 ofproto->frag_handling = next;
2340 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2342 ofputil_frag_handling_to_string(next));
2346 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2347 ofconn_set_invalid_ttl_to_controller(ofconn,
2348 (oh->version < OFP13_VERSION
2349 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2351 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2356 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2357 * error message code for the caller to propagate upward. Otherwise, returns
2360 * The log message mentions 'msg_type'. */
2362 reject_slave_controller(struct ofconn *ofconn)
2364 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2365 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2366 return OFPERR_OFPBRC_EPERM;
2373 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2375 struct ofproto *p = ofconn_get_ofproto(ofconn);
2376 struct ofputil_packet_out po;
2377 struct ofpbuf *payload;
2378 uint64_t ofpacts_stub[1024 / 8];
2379 struct ofpbuf ofpacts;
2381 union flow_in_port in_port_;
2384 COVERAGE_INC(ofproto_packet_out);
2386 error = reject_slave_controller(ofconn);
2391 /* Decode message. */
2392 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2393 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2395 goto exit_free_ofpacts;
2397 if (ofp_to_u16(po.in_port) >= ofp_to_u16(p->max_ports)
2398 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2399 error = OFPERR_OFPBRC_BAD_PORT;
2400 goto exit_free_ofpacts;
2405 if (po.buffer_id != UINT32_MAX) {
2406 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2407 if (error || !payload) {
2408 goto exit_free_ofpacts;
2411 payload = xmalloc(sizeof *payload);
2412 ofpbuf_use_const(payload, po.packet, po.packet_len);
2415 /* Verify actions against packet, then send packet if successful. */
2416 in_port_.ofp_port = po.in_port;
2417 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2418 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2420 error = p->ofproto_class->packet_out(p, payload, &flow,
2421 po.ofpacts, po.ofpacts_len);
2423 ofpbuf_delete(payload);
2426 ofpbuf_uninit(&ofpacts);
2432 update_port_config(struct ofport *port,
2433 enum ofputil_port_config config,
2434 enum ofputil_port_config mask)
2436 enum ofputil_port_config old_config = port->pp.config;
2437 enum ofputil_port_config toggle;
2439 toggle = (config ^ port->pp.config) & mask;
2440 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2441 if (config & OFPUTIL_PC_PORT_DOWN) {
2442 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2444 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2446 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2449 port->pp.config ^= toggle;
2450 if (port->pp.config != old_config) {
2451 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2456 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2458 struct ofproto *p = ofconn_get_ofproto(ofconn);
2459 struct ofputil_port_mod pm;
2460 struct ofport *port;
2463 error = reject_slave_controller(ofconn);
2468 error = ofputil_decode_port_mod(oh, &pm);
2473 port = ofproto_get_port(p, pm.port_no);
2475 return OFPERR_OFPPMFC_BAD_PORT;
2476 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2477 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2479 update_port_config(port, pm.config, pm.mask);
2481 netdev_set_advertisements(port->netdev, pm.advertise);
2488 handle_desc_stats_request(struct ofconn *ofconn,
2489 const struct ofp_header *request)
2491 static const char *default_mfr_desc = "Nicira, Inc.";
2492 static const char *default_hw_desc = "Open vSwitch";
2493 static const char *default_sw_desc = VERSION;
2494 static const char *default_serial_desc = "None";
2495 static const char *default_dp_desc = "None";
2497 struct ofproto *p = ofconn_get_ofproto(ofconn);
2498 struct ofp_desc_stats *ods;
2501 msg = ofpraw_alloc_stats_reply(request, 0);
2502 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2503 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2504 sizeof ods->mfr_desc);
2505 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2506 sizeof ods->hw_desc);
2507 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2508 sizeof ods->sw_desc);
2509 ovs_strlcpy(ods->serial_num,
2510 p->serial_desc ? p->serial_desc : default_serial_desc,
2511 sizeof ods->serial_num);
2512 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2513 sizeof ods->dp_desc);
2514 ofconn_send_reply(ofconn, msg);
2520 handle_table_stats_request(struct ofconn *ofconn,
2521 const struct ofp_header *request)
2523 struct ofproto *p = ofconn_get_ofproto(ofconn);
2524 struct ofp12_table_stats *ots;
2529 /* Set up default values.
2531 * ofp12_table_stats is used as a generic structure as
2532 * it is able to hold all the fields for ofp10_table_stats
2533 * and ofp11_table_stats (and of course itself).
2535 ots = xcalloc(p->n_tables, sizeof *ots);
2536 for (i = 0; i < p->n_tables; i++) {
2537 ots[i].table_id = i;
2538 sprintf(ots[i].name, "table%zu", i);
2539 ots[i].match = htonll(OFPXMT12_MASK);
2540 ots[i].wildcards = htonll(OFPXMT12_MASK);
2541 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2542 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2543 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2544 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2545 ots[i].metadata_match = htonll(UINT64_MAX);
2546 ots[i].metadata_write = htonll(UINT64_MAX);
2547 ots[i].instructions = htonl(OFPIT11_ALL);
2548 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2549 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2550 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2553 p->ofproto_class->get_tables(p, ots);
2555 /* Post-process the tables, dropping hidden tables. */
2556 n_tables = p->n_tables;
2557 for (i = 0; i < p->n_tables; i++) {
2558 const struct oftable *table = &p->tables[i];
2560 if (table->flags & OFTABLE_HIDDEN) {
2566 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2569 if (table->max_flows < ntohl(ots[i].max_entries)) {
2570 ots[i].max_entries = htonl(table->max_flows);
2574 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2575 ofconn_send_reply(ofconn, msg);
2583 append_port_stat(struct ofport *port, struct list *replies)
2585 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2587 calc_duration(port->created, time_msec(),
2588 &ops.duration_sec, &ops.duration_nsec);
2590 /* Intentionally ignore return value, since errors will set
2591 * 'stats' to all-1s, which is correct for OpenFlow, and
2592 * netdev_get_stats() will log errors. */
2593 ofproto_port_get_stats(port, &ops.stats);
2595 ofputil_append_port_stat(replies, &ops);
2599 handle_port_stats_request(struct ofconn *ofconn,
2600 const struct ofp_header *request)
2602 struct ofproto *p = ofconn_get_ofproto(ofconn);
2603 struct ofport *port;
2604 struct list replies;
2608 error = ofputil_decode_port_stats_request(request, &port_no);
2613 ofpmp_init(&replies, request);
2614 if (port_no != OFPP_ANY) {
2615 port = ofproto_get_port(p, port_no);
2617 append_port_stat(port, &replies);
2620 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2621 append_port_stat(port, &replies);
2625 ofconn_send_replies(ofconn, &replies);
2630 handle_port_desc_stats_request(struct ofconn *ofconn,
2631 const struct ofp_header *request)
2633 struct ofproto *p = ofconn_get_ofproto(ofconn);
2634 enum ofp_version version;
2635 struct ofport *port;
2636 struct list replies;
2638 ofpmp_init(&replies, request);
2640 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2641 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2642 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2645 ofconn_send_replies(ofconn, &replies);
2650 hash_cookie(ovs_be64 cookie)
2652 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2653 (OVS_FORCE uint64_t)cookie);
2657 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2659 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2660 hash_cookie(rule->flow_cookie));
2664 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2666 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2670 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2671 ovs_be64 new_cookie)
2673 if (new_cookie != rule->flow_cookie) {
2674 cookies_remove(ofproto, rule);
2676 rule->flow_cookie = new_cookie;
2678 cookies_insert(ofproto, rule);
2683 calc_duration(long long int start, long long int now,
2684 uint32_t *sec, uint32_t *nsec)
2686 long long int msecs = now - start;
2687 *sec = msecs / 1000;
2688 *nsec = (msecs % 1000) * (1000 * 1000);
2691 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2692 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2694 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2696 return (table_id == 0xff || table_id < ofproto->n_tables
2698 : OFPERR_OFPBRC_BAD_TABLE_ID);
2702 static struct oftable *
2703 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2705 struct oftable *table;
2707 for (table = &ofproto->tables[table_id];
2708 table < &ofproto->tables[ofproto->n_tables];
2710 if (!(table->flags & OFTABLE_HIDDEN)) {
2718 static struct oftable *
2719 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2721 if (table_id == 0xff) {
2722 return next_visible_table(ofproto, 0);
2723 } else if (table_id < ofproto->n_tables) {
2724 return &ofproto->tables[table_id];
2730 static struct oftable *
2731 next_matching_table(const struct ofproto *ofproto,
2732 const struct oftable *table, uint8_t table_id)
2734 return (table_id == 0xff
2735 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2739 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2741 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2742 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2744 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2745 * only once, for that table. (This can be used to access tables marked
2748 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2749 * entered at all. (Perhaps you should have validated TABLE_ID with
2750 * check_table_id().)
2752 * All parameters are evaluated multiple times.
2754 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2755 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2757 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2759 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2760 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2761 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2764 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2765 * to 'out_port' are included.
2767 * Hidden rules are always omitted.
2769 * Returns 0 on success, otherwise an OpenFlow error code. */
2771 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2772 const struct match *match,
2773 ovs_be64 cookie, ovs_be64 cookie_mask,
2774 ofp_port_t out_port, struct list *rules)
2776 struct oftable *table;
2780 error = check_table_id(ofproto, table_id);
2786 cls_rule_init(&cr, match, 0);
2788 if (cookie_mask == htonll(UINT64_MAX)) {
2791 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2792 &ofproto->cookies) {
2793 if (table_id != rule->table_id && table_id != 0xff) {
2796 if (ofproto_rule_is_hidden(rule)) {
2799 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2800 if (rule->pending) {
2801 error = OFPROTO_POSTPONE;
2804 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2805 && ofproto_rule_has_out_port(rule, out_port)) {
2806 list_push_back(rules, &rule->ofproto_node);
2813 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2814 struct cls_cursor cursor;
2817 cls_cursor_init(&cursor, &table->cls, &cr);
2818 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2819 if (rule->pending) {
2820 error = OFPROTO_POSTPONE;
2823 if (!ofproto_rule_is_hidden(rule)
2824 && ofproto_rule_has_out_port(rule, out_port)
2825 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2826 list_push_back(rules, &rule->ofproto_node);
2832 cls_rule_destroy(&cr);
2836 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2837 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2838 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2841 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2842 * to 'out_port' are included.
2844 * Hidden rules are always omitted.
2846 * Returns 0 on success, otherwise an OpenFlow error code. */
2848 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2849 const struct match *match, unsigned int priority,
2850 ovs_be64 cookie, ovs_be64 cookie_mask,
2851 ofp_port_t out_port, struct list *rules)
2853 struct oftable *table;
2857 error = check_table_id(ofproto, table_id);
2863 cls_rule_init(&cr, match, priority);
2865 if (cookie_mask == htonll(UINT64_MAX)) {
2868 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2869 &ofproto->cookies) {
2870 if (table_id != rule->table_id && table_id != 0xff) {
2873 if (ofproto_rule_is_hidden(rule)) {
2876 if (cls_rule_equal(&rule->cr, &cr)) {
2877 if (rule->pending) {
2878 error = OFPROTO_POSTPONE;
2881 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2882 && ofproto_rule_has_out_port(rule, out_port)) {
2883 list_push_back(rules, &rule->ofproto_node);
2890 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2893 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2896 if (rule->pending) {
2897 error = OFPROTO_POSTPONE;
2900 if (!ofproto_rule_is_hidden(rule)
2901 && ofproto_rule_has_out_port(rule, out_port)
2902 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2903 list_push_back(rules, &rule->ofproto_node);
2909 cls_rule_destroy(&cr);
2913 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2914 * forced into the range of a uint16_t. */
2916 age_secs(long long int age_ms)
2918 return (age_ms < 0 ? 0
2919 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2920 : (unsigned int) age_ms / 1000);
2924 handle_flow_stats_request(struct ofconn *ofconn,
2925 const struct ofp_header *request)
2927 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2928 struct ofputil_flow_stats_request fsr;
2929 struct list replies;
2934 error = ofputil_decode_flow_stats_request(&fsr, request);
2939 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2940 fsr.cookie, fsr.cookie_mask,
2941 fsr.out_port, &rules);
2946 ofpmp_init(&replies, request);
2947 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2948 long long int now = time_msec();
2949 struct ofputil_flow_stats fs;
2951 minimatch_expand(&rule->cr.match, &fs.match);
2952 fs.priority = rule->cr.priority;
2953 fs.cookie = rule->flow_cookie;
2954 fs.table_id = rule->table_id;
2955 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
2956 fs.idle_timeout = rule->idle_timeout;
2957 fs.hard_timeout = rule->hard_timeout;
2958 fs.idle_age = age_secs(now - rule->used);
2959 fs.hard_age = age_secs(now - rule->modified);
2960 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2962 fs.ofpacts = rule->ofpacts;
2963 fs.ofpacts_len = rule->ofpacts_len;
2965 if (rule->send_flow_removed) {
2966 fs.flags |= OFPFF_SEND_FLOW_REM;
2967 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2968 and OFPFF13_NO_BYT_COUNTS */
2970 ofputil_append_flow_stats_reply(&fs, &replies);
2972 ofconn_send_replies(ofconn, &replies);
2978 flow_stats_ds(struct rule *rule, struct ds *results)
2980 uint64_t packet_count, byte_count;
2982 rule->ofproto->ofproto_class->rule_get_stats(rule,
2983 &packet_count, &byte_count);
2985 if (rule->table_id != 0) {
2986 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2988 ds_put_format(results, "duration=%llds, ",
2989 (time_msec() - rule->created) / 1000);
2990 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2991 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2992 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2993 cls_rule_format(&rule->cr, results);
2994 ds_put_char(results, ',');
2995 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2996 ds_put_cstr(results, "\n");
2999 /* Adds a pretty-printed description of all flows to 'results', including
3000 * hidden flows (e.g., set up by in-band control). */
3002 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3004 struct oftable *table;
3006 OFPROTO_FOR_EACH_TABLE (table, p) {
3007 struct cls_cursor cursor;
3010 cls_cursor_init(&cursor, &table->cls, NULL);
3011 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3012 flow_stats_ds(rule, results);
3017 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3018 * '*engine_type' and '*engine_id', respectively. */
3020 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3021 uint8_t *engine_type, uint8_t *engine_id)
3023 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3026 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3027 * true if the port's CFM status was successfully stored into '*status'.
3028 * Returns false if the port did not have CFM configured, in which case
3029 * '*status' is indeterminate.
3031 * The caller must provide and owns '*status', but it does not own and must not
3032 * modify or free the array returned in 'status->rmps'. */
3034 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3035 struct ofproto_cfm_status *status)
3037 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3039 && ofproto->ofproto_class->get_cfm_status
3040 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3044 handle_aggregate_stats_request(struct ofconn *ofconn,
3045 const struct ofp_header *oh)
3047 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3048 struct ofputil_flow_stats_request request;
3049 struct ofputil_aggregate_stats stats;
3050 bool unknown_packets, unknown_bytes;
3051 struct ofpbuf *reply;
3056 error = ofputil_decode_flow_stats_request(&request, oh);
3061 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3062 request.cookie, request.cookie_mask,
3063 request.out_port, &rules);
3068 memset(&stats, 0, sizeof stats);
3069 unknown_packets = unknown_bytes = false;
3070 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3071 uint64_t packet_count;
3072 uint64_t byte_count;
3074 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3077 if (packet_count == UINT64_MAX) {
3078 unknown_packets = true;
3080 stats.packet_count += packet_count;
3083 if (byte_count == UINT64_MAX) {
3084 unknown_bytes = true;
3086 stats.byte_count += byte_count;
3091 if (unknown_packets) {
3092 stats.packet_count = UINT64_MAX;
3094 if (unknown_bytes) {
3095 stats.byte_count = UINT64_MAX;
3098 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3099 ofconn_send_reply(ofconn, reply);
3104 struct queue_stats_cbdata {
3105 struct ofport *ofport;
3106 struct list replies;
3110 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3111 const struct netdev_queue_stats *stats)
3114 struct ofputil_queue_stats oqs = {
3115 .port_no = cbdata->ofport->pp.port_no,
3116 .queue_id = queue_id,
3119 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3123 handle_queue_stats_dump_cb(uint32_t queue_id,
3124 struct netdev_queue_stats *stats,
3127 struct queue_stats_cbdata *cbdata = cbdata_;
3129 put_queue_stats(cbdata, queue_id, stats);
3133 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3134 struct queue_stats_cbdata *cbdata)
3136 cbdata->ofport = port;
3137 if (queue_id == OFPQ_ALL) {
3138 netdev_dump_queue_stats(port->netdev,
3139 handle_queue_stats_dump_cb, cbdata);
3141 struct netdev_queue_stats stats;
3143 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3144 put_queue_stats(cbdata, queue_id, &stats);
3146 return OFPERR_OFPQOFC_BAD_QUEUE;
3153 handle_queue_stats_request(struct ofconn *ofconn,
3154 const struct ofp_header *rq)
3156 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3157 struct queue_stats_cbdata cbdata;
3158 struct ofport *port;
3160 struct ofputil_queue_stats_request oqsr;
3162 COVERAGE_INC(ofproto_queue_req);
3164 ofpmp_init(&cbdata.replies, rq);
3166 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3171 if (oqsr.port_no == OFPP_ANY) {
3172 error = OFPERR_OFPQOFC_BAD_QUEUE;
3173 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3174 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3179 port = ofproto_get_port(ofproto, oqsr.port_no);
3181 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3182 : OFPERR_OFPQOFC_BAD_PORT);
3185 ofconn_send_replies(ofconn, &cbdata.replies);
3187 ofpbuf_list_delete(&cbdata.replies);
3194 is_flow_deletion_pending(const struct ofproto *ofproto,
3195 const struct cls_rule *cls_rule,
3198 if (!hmap_is_empty(&ofproto->deletions)) {
3199 struct ofoperation *op;
3201 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3202 cls_rule_hash(cls_rule, table_id),
3203 &ofproto->deletions) {
3204 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3213 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3214 * in which no matching flow already exists in the flow table.
3216 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3217 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3218 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3219 * initiated now but may be retried later.
3221 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3222 * ownership remains with the caller.
3224 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3227 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3228 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3230 struct oftable *table;
3231 struct ofopgroup *group;
3232 struct rule *victim;
3237 error = check_table_id(ofproto, fm->table_id);
3243 if (fm->table_id == 0xff) {
3244 if (ofproto->ofproto_class->rule_choose_table) {
3245 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3251 ovs_assert(table_id < ofproto->n_tables);
3255 } else if (fm->table_id < ofproto->n_tables) {
3256 table_id = fm->table_id;
3258 return OFPERR_OFPBRC_BAD_TABLE_ID;
3261 table = &ofproto->tables[table_id];
3263 if (table->flags & OFTABLE_READONLY) {
3264 return OFPERR_OFPBRC_EPERM;
3267 /* Allocate new rule and initialize classifier rule. */
3268 rule = ofproto->ofproto_class->rule_alloc();
3270 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3271 ofproto->name, strerror(error));
3274 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3276 /* Serialize against pending deletion. */
3277 if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) {
3278 cls_rule_destroy(&rule->cr);
3279 ofproto->ofproto_class->rule_dealloc(rule);
3280 return OFPROTO_POSTPONE;
3283 /* Check for overlap, if requested. */
3284 if (fm->flags & OFPFF_CHECK_OVERLAP
3285 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3286 cls_rule_destroy(&rule->cr);
3287 ofproto->ofproto_class->rule_dealloc(rule);
3288 return OFPERR_OFPFMFC_OVERLAP;
3291 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3293 rule->ofproto = ofproto;
3294 rule->pending = NULL;
3295 rule->flow_cookie = fm->new_cookie;
3296 rule->created = rule->modified = rule->used = time_msec();
3297 rule->idle_timeout = fm->idle_timeout;
3298 rule->hard_timeout = fm->hard_timeout;
3299 rule->table_id = table - ofproto->tables;
3300 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3301 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3302 and OFPFF13_NO_BYT_COUNTS */
3303 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3304 rule->ofpacts_len = fm->ofpacts_len;
3305 rule->evictable = true;
3306 rule->eviction_group = NULL;
3307 list_init(&rule->expirable);
3308 rule->monitor_flags = 0;
3309 rule->add_seqno = 0;
3310 rule->modify_seqno = 0;
3312 /* Insert new rule. */
3313 victim = oftable_replace_rule(rule);
3314 if (victim && !rule_is_modifiable(victim)) {
3315 error = OFPERR_OFPBRC_EPERM;
3316 } else if (victim && victim->pending) {
3317 error = OFPROTO_POSTPONE;
3319 struct ofoperation *op;
3322 if (classifier_count(&table->cls) > table->max_flows) {
3325 was_evictable = rule->evictable;
3326 rule->evictable = false;
3327 evict = choose_rule_to_evict(table);
3328 rule->evictable = was_evictable;
3331 error = OFPERR_OFPFMFC_TABLE_FULL;
3333 } else if (evict->pending) {
3334 error = OFPROTO_POSTPONE;
3341 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3342 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3343 op->victim = victim;
3345 error = ofproto->ofproto_class->rule_construct(rule);
3347 op->group->n_running--;
3348 ofoperation_destroy(rule->pending);
3350 delete_flow__(evict, group, OFPRR_EVICTION);
3352 ofopgroup_submit(group);
3356 /* Back out if an error occurred. */
3358 oftable_substitute_rule(rule, victim);
3359 ofproto_rule_destroy__(rule);
3364 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3366 /* Modifies the rules listed in 'rules', changing their actions to match those
3369 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3372 * Returns 0 on success, otherwise an OpenFlow error code. */
3374 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3375 const struct ofputil_flow_mod *fm,
3376 const struct ofp_header *request, struct list *rules)
3378 struct ofopgroup *group;
3382 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3383 error = OFPERR_OFPBRC_EPERM;
3384 LIST_FOR_EACH (rule, ofproto_node, rules) {
3385 struct ofoperation *op;
3386 bool actions_changed;
3388 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3390 if (rule_is_modifiable(rule)) {
3391 /* At least one rule is modifiable, don't report EPERM error. */
3397 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3398 rule->ofpacts, rule->ofpacts_len);
3400 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3402 if (fm->new_cookie != htonll(UINT64_MAX)) {
3403 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3405 if (actions_changed) {
3406 op->ofpacts = rule->ofpacts;
3407 op->ofpacts_len = rule->ofpacts_len;
3408 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3409 rule->ofpacts_len = fm->ofpacts_len;
3410 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3412 ofoperation_complete(op, 0);
3415 ofopgroup_submit(group);
3421 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3422 const struct ofputil_flow_mod *fm,
3423 const struct ofp_header *request)
3425 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3428 return add_flow(ofproto, ofconn, fm, request);
3431 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3434 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3437 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3438 const struct ofputil_flow_mod *fm,
3439 const struct ofp_header *request)
3444 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3445 fm->cookie, fm->cookie_mask,
3449 } else if (list_is_empty(&rules)) {
3450 return modify_flows_add(ofproto, ofconn, fm, request);
3452 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3456 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3459 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3462 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3463 const struct ofputil_flow_mod *fm,
3464 const struct ofp_header *request)
3469 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3470 fm->priority, fm->cookie, fm->cookie_mask,
3475 } else if (list_is_empty(&rules)) {
3476 return modify_flows_add(ofproto, ofconn, fm, request);
3478 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3479 fm, request, &rules)
3484 /* OFPFC_DELETE implementation. */
3487 delete_flow__(struct rule *rule, struct ofopgroup *group,
3488 enum ofp_flow_removed_reason reason)
3490 struct ofproto *ofproto = rule->ofproto;
3492 ofproto_rule_send_removed(rule, reason);
3494 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3495 oftable_remove_rule(rule);
3496 ofproto->ofproto_class->rule_destruct(rule);
3499 /* Deletes the rules listed in 'rules'.
3501 * Returns 0 on success, otherwise an OpenFlow error code. */
3503 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3504 const struct ofp_header *request, struct list *rules,
3505 enum ofp_flow_removed_reason reason)
3507 struct rule *rule, *next;
3508 struct ofopgroup *group;
3510 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3511 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3512 delete_flow__(rule, group, reason);
3514 ofopgroup_submit(group);
3519 /* Implements OFPFC_DELETE. */
3521 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3522 const struct ofputil_flow_mod *fm,
3523 const struct ofp_header *request)
3528 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3529 fm->cookie, fm->cookie_mask,
3530 fm->out_port, &rules);
3531 return (error ? error
3532 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3533 &rules, OFPRR_DELETE)
3537 /* Implements OFPFC_DELETE_STRICT. */
3539 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3540 const struct ofputil_flow_mod *fm,
3541 const struct ofp_header *request)
3546 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3547 fm->priority, fm->cookie, fm->cookie_mask,
3548 fm->out_port, &rules);
3549 return (error ? error
3550 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3557 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3559 struct ofputil_flow_removed fr;
3561 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3565 minimatch_expand(&rule->cr.match, &fr.match);
3566 fr.priority = rule->cr.priority;
3567 fr.cookie = rule->flow_cookie;
3569 fr.table_id = rule->table_id;
3570 calc_duration(rule->created, time_msec(),
3571 &fr.duration_sec, &fr.duration_nsec);
3572 fr.idle_timeout = rule->idle_timeout;
3573 fr.hard_timeout = rule->hard_timeout;
3574 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3577 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3581 ofproto_rule_update_used(struct rule *rule, long long int used)
3583 if (used > rule->used) {
3584 struct eviction_group *evg = rule->eviction_group;
3588 heap_change(&evg->rules, &rule->evg_node,
3589 rule_eviction_priority(rule));
3594 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3595 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3598 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3601 * ofproto implementation ->run() functions should use this function to expire
3602 * OpenFlow flows. */
3604 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3606 struct ofproto *ofproto = rule->ofproto;
3607 struct ofopgroup *group;
3609 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3611 ofproto_rule_send_removed(rule, reason);
3613 group = ofopgroup_create_unattached(ofproto);
3614 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3615 oftable_remove_rule(rule);
3616 ofproto->ofproto_class->rule_destruct(rule);
3617 ofopgroup_submit(group);
3621 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3623 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3624 struct ofputil_flow_mod fm;
3625 uint64_t ofpacts_stub[1024 / 8];
3626 struct ofpbuf ofpacts;
3630 error = reject_slave_controller(ofconn);
3635 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3636 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3639 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3640 &fm.match.flow, ofproto->max_ports);
3643 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3646 goto exit_free_ofpacts;
3649 /* Record the operation for logging a summary report. */
3650 switch (fm.command) {
3656 case OFPFC_MODIFY_STRICT:
3657 ofproto->n_modify++;
3661 case OFPFC_DELETE_STRICT:
3662 ofproto->n_delete++;
3667 if (ofproto->next_op_report == LLONG_MAX) {
3668 ofproto->first_op = now;
3669 ofproto->next_op_report = MAX(now + 10 * 1000,
3670 ofproto->op_backoff);
3671 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3673 ofproto->last_op = now;
3676 ofpbuf_uninit(&ofpacts);
3682 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3683 const struct ofputil_flow_mod *fm,
3684 const struct ofp_header *oh)
3686 if (ofproto->n_pending >= 50) {
3687 ovs_assert(!list_is_empty(&ofproto->pending));
3688 return OFPROTO_POSTPONE;
3691 switch (fm->command) {
3693 return add_flow(ofproto, ofconn, fm, oh);
3696 return modify_flows_loose(ofproto, ofconn, fm, oh);
3698 case OFPFC_MODIFY_STRICT:
3699 return modify_flow_strict(ofproto, ofconn, fm, oh);
3702 return delete_flows_loose(ofproto, ofconn, fm, oh);
3704 case OFPFC_DELETE_STRICT:
3705 return delete_flow_strict(ofproto, ofconn, fm, oh);
3708 if (fm->command > 0xff) {
3709 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3710 "flow_mod_table_id extension is not enabled",
3713 return OFPERR_OFPFMFC_BAD_COMMAND;
3718 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3720 struct ofputil_role_request request;
3721 struct ofputil_role_request reply;
3725 error = ofputil_decode_role_message(oh, &request);
3730 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3731 if (ofconn_get_role(ofconn) != request.role
3732 && ofconn_has_pending_opgroups(ofconn)) {
3733 return OFPROTO_POSTPONE;
3736 if (request.have_generation_id
3737 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3738 return OFPERR_OFPRRFC_STALE;
3741 ofconn_set_role(ofconn, request.role);
3744 reply.role = ofconn_get_role(ofconn);
3745 reply.have_generation_id = ofconn_get_master_election_id(
3746 ofconn, &reply.generation_id);
3747 buf = ofputil_encode_role_reply(oh, &reply);
3748 ofconn_send_reply(ofconn, buf);
3754 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3755 const struct ofp_header *oh)
3757 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3758 enum ofputil_protocol cur, next;
3760 cur = ofconn_get_protocol(ofconn);
3761 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3762 ofconn_set_protocol(ofconn, next);
3768 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3770 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3771 enum ofputil_protocol cur, next;
3772 enum ofputil_protocol next_base;
3774 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3776 return OFPERR_OFPBRC_EPERM;
3779 cur = ofconn_get_protocol(ofconn);
3780 next = ofputil_protocol_set_base(cur, next_base);
3781 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3782 /* Avoid sending async messages in surprising protocol. */
3783 return OFPROTO_POSTPONE;
3786 ofconn_set_protocol(ofconn, next);
3791 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3792 const struct ofp_header *oh)
3794 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3797 format = ntohl(msg->format);
3798 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3799 return OFPERR_OFPBRC_EPERM;
3802 if (format != ofconn_get_packet_in_format(ofconn)
3803 && ofconn_has_pending_opgroups(ofconn)) {
3804 /* Avoid sending async message in surprsing packet in format. */
3805 return OFPROTO_POSTPONE;
3808 ofconn_set_packet_in_format(ofconn, format);
3813 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3815 const struct nx_async_config *msg = ofpmsg_body(oh);
3816 uint32_t master[OAM_N_TYPES];
3817 uint32_t slave[OAM_N_TYPES];
3819 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3820 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3821 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3823 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3824 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3825 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3827 ofconn_set_async_config(ofconn, master, slave);
3828 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3829 !ofconn_get_miss_send_len(ofconn)) {
3830 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3837 handle_nxt_set_controller_id(struct ofconn *ofconn,
3838 const struct ofp_header *oh)
3840 const struct nx_controller_id *nci = ofpmsg_body(oh);
3842 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3843 return OFPERR_NXBRC_MUST_BE_ZERO;
3846 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3851 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3855 if (ofconn_has_pending_opgroups(ofconn)) {
3856 return OFPROTO_POSTPONE;
3859 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3860 ? OFPRAW_OFPT10_BARRIER_REPLY
3861 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3862 ofconn_send_reply(ofconn, buf);
3867 ofproto_compose_flow_refresh_update(const struct rule *rule,
3868 enum nx_flow_monitor_flags flags,
3871 struct ofoperation *op = rule->pending;
3872 struct ofputil_flow_update fu;
3875 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3876 /* We'll report the final flow when the operation completes. Reporting
3877 * it now would cause a duplicate report later. */
3881 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3882 ? NXFME_ADDED : NXFME_MODIFIED);
3884 fu.idle_timeout = rule->idle_timeout;
3885 fu.hard_timeout = rule->hard_timeout;
3886 fu.table_id = rule->table_id;
3887 fu.cookie = rule->flow_cookie;
3888 minimatch_expand(&rule->cr.match, &match);
3890 fu.priority = rule->cr.priority;
3891 if (!(flags & NXFMF_ACTIONS)) {
3895 fu.ofpacts = rule->ofpacts;
3896 fu.ofpacts_len = rule->ofpacts_len;
3898 /* An operation is in progress. Use the previous version of the flow's
3899 * actions, so that when the operation commits we report the change. */
3901 case OFOPERATION_ADD:
3902 /* We already verified that there was a victim. */
3903 fu.ofpacts = op->victim->ofpacts;
3904 fu.ofpacts_len = op->victim->ofpacts_len;
3907 case OFOPERATION_MODIFY:
3909 fu.ofpacts = op->ofpacts;
3910 fu.ofpacts_len = op->ofpacts_len;
3912 fu.ofpacts = rule->ofpacts;
3913 fu.ofpacts_len = rule->ofpacts_len;
3917 case OFOPERATION_DELETE:
3918 fu.ofpacts = rule->ofpacts;
3919 fu.ofpacts_len = rule->ofpacts_len;
3927 if (list_is_empty(msgs)) {
3928 ofputil_start_flow_update(msgs);
3930 ofputil_append_flow_update(&fu, msgs);
3934 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3938 LIST_FOR_EACH (rule, ofproto_node, rules) {
3939 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3940 rule->monitor_flags = 0;
3942 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3947 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3948 struct rule *rule, uint64_t seqno,
3951 enum nx_flow_monitor_flags update;
3953 if (ofproto_rule_is_hidden(rule)) {
3958 ? ofoperation_has_out_port(rule->pending, m->out_port)
3959 : ofproto_rule_has_out_port(rule, m->out_port))) {
3964 if (rule->add_seqno > seqno) {
3965 update = NXFMF_ADD | NXFMF_MODIFY;
3966 } else if (rule->modify_seqno > seqno) {
3967 update = NXFMF_MODIFY;
3972 if (!(m->flags & update)) {
3976 update = NXFMF_INITIAL;
3979 if (!rule->monitor_flags) {
3980 list_push_back(rules, &rule->ofproto_node);
3982 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3986 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3990 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3991 const struct ofoperation *op;
3992 const struct oftable *table;
3993 struct cls_rule target;
3995 cls_rule_init_from_minimatch(&target, &m->match, 0);
3996 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3997 struct cls_cursor cursor;
4000 cls_cursor_init(&cursor, &table->cls, &target);
4001 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4002 ovs_assert(!rule->pending); /* XXX */
4003 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4007 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4008 struct rule *rule = op->rule;
4010 if (((m->table_id == 0xff
4011 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4012 : m->table_id == rule->table_id))
4013 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4014 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4017 cls_rule_destroy(&target);
4021 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4024 if (m->flags & NXFMF_INITIAL) {
4025 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4030 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4031 uint64_t seqno, struct list *rules)
4033 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4037 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4039 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4040 struct ofmonitor **monitors;
4041 size_t n_monitors, allocated_monitors;
4042 struct list replies;
4049 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4051 n_monitors = allocated_monitors = 0;
4053 struct ofputil_flow_monitor_request request;
4054 struct ofmonitor *m;
4057 retval = ofputil_decode_flow_monitor_request(&request, &b);
4058 if (retval == EOF) {
4060 } else if (retval) {
4065 if (request.table_id != 0xff
4066 && request.table_id >= ofproto->n_tables) {
4067 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4071 error = ofmonitor_create(&request, ofconn, &m);
4076 if (n_monitors >= allocated_monitors) {
4077 monitors = x2nrealloc(monitors, &allocated_monitors,
4080 monitors[n_monitors++] = m;
4084 for (i = 0; i < n_monitors; i++) {
4085 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4088 ofpmp_init(&replies, oh);
4089 ofmonitor_compose_refresh_updates(&rules, &replies);
4090 ofconn_send_replies(ofconn, &replies);
4097 for (i = 0; i < n_monitors; i++) {
4098 ofmonitor_destroy(monitors[i]);
4105 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4107 struct ofmonitor *m;
4110 id = ofputil_decode_flow_monitor_cancel(oh);
4111 m = ofmonitor_lookup(ofconn, id);
4113 return OFPERR_NXBRC_FM_BAD_ID;
4116 ofmonitor_destroy(m);
4121 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4123 const struct ofp_header *oh = msg->data;
4127 error = ofptype_decode(&type, oh);
4133 /* OpenFlow requests. */
4134 case OFPTYPE_ECHO_REQUEST:
4135 return handle_echo_request(ofconn, oh);
4137 case OFPTYPE_FEATURES_REQUEST:
4138 return handle_features_request(ofconn, oh);
4140 case OFPTYPE_GET_CONFIG_REQUEST:
4141 return handle_get_config_request(ofconn, oh);
4143 case OFPTYPE_SET_CONFIG:
4144 return handle_set_config(ofconn, oh);
4146 case OFPTYPE_PACKET_OUT:
4147 return handle_packet_out(ofconn, oh);
4149 case OFPTYPE_PORT_MOD:
4150 return handle_port_mod(ofconn, oh);
4152 case OFPTYPE_FLOW_MOD:
4153 return handle_flow_mod(ofconn, oh);
4155 case OFPTYPE_BARRIER_REQUEST:
4156 return handle_barrier_request(ofconn, oh);
4158 case OFPTYPE_ROLE_REQUEST:
4159 return handle_role_request(ofconn, oh);
4161 /* OpenFlow replies. */
4162 case OFPTYPE_ECHO_REPLY:
4165 /* Nicira extension requests. */
4166 case OFPTYPE_FLOW_MOD_TABLE_ID:
4167 return handle_nxt_flow_mod_table_id(ofconn, oh);
4169 case OFPTYPE_SET_FLOW_FORMAT:
4170 return handle_nxt_set_flow_format(ofconn, oh);
4172 case OFPTYPE_SET_PACKET_IN_FORMAT:
4173 return handle_nxt_set_packet_in_format(ofconn, oh);
4175 case OFPTYPE_SET_CONTROLLER_ID:
4176 return handle_nxt_set_controller_id(ofconn, oh);
4178 case OFPTYPE_FLOW_AGE:
4179 /* Nothing to do. */
4182 case OFPTYPE_FLOW_MONITOR_CANCEL:
4183 return handle_flow_monitor_cancel(ofconn, oh);
4185 case OFPTYPE_SET_ASYNC_CONFIG:
4186 return handle_nxt_set_async_config(ofconn, oh);
4188 /* Statistics requests. */
4189 case OFPTYPE_DESC_STATS_REQUEST:
4190 return handle_desc_stats_request(ofconn, oh);
4192 case OFPTYPE_FLOW_STATS_REQUEST:
4193 return handle_flow_stats_request(ofconn, oh);
4195 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4196 return handle_aggregate_stats_request(ofconn, oh);
4198 case OFPTYPE_TABLE_STATS_REQUEST:
4199 return handle_table_stats_request(ofconn, oh);
4201 case OFPTYPE_PORT_STATS_REQUEST:
4202 return handle_port_stats_request(ofconn, oh);
4204 case OFPTYPE_QUEUE_STATS_REQUEST:
4205 return handle_queue_stats_request(ofconn, oh);
4207 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4208 return handle_port_desc_stats_request(ofconn, oh);
4210 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4211 return handle_flow_monitor_request(ofconn, oh);
4213 /* FIXME: Change the following once they are implemented: */
4214 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4215 case OFPTYPE_GET_ASYNC_REQUEST:
4216 case OFPTYPE_METER_MOD:
4217 case OFPTYPE_GROUP_REQUEST:
4218 case OFPTYPE_GROUP_DESC_REQUEST:
4219 case OFPTYPE_GROUP_FEATURES_REQUEST:
4220 case OFPTYPE_METER_REQUEST:
4221 case OFPTYPE_METER_CONFIG_REQUEST:
4222 case OFPTYPE_METER_FEATURES_REQUEST:
4223 case OFPTYPE_TABLE_FEATURES_REQUEST:
4224 return OFPERR_OFPBRC_BAD_TYPE;
4228 case OFPTYPE_FEATURES_REPLY:
4229 case OFPTYPE_GET_CONFIG_REPLY:
4230 case OFPTYPE_PACKET_IN:
4231 case OFPTYPE_FLOW_REMOVED:
4232 case OFPTYPE_PORT_STATUS:
4233 case OFPTYPE_BARRIER_REPLY:
4234 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4235 case OFPTYPE_DESC_STATS_REPLY:
4236 case OFPTYPE_FLOW_STATS_REPLY:
4237 case OFPTYPE_QUEUE_STATS_REPLY:
4238 case OFPTYPE_PORT_STATS_REPLY:
4239 case OFPTYPE_TABLE_STATS_REPLY:
4240 case OFPTYPE_AGGREGATE_STATS_REPLY:
4241 case OFPTYPE_PORT_DESC_STATS_REPLY:
4242 case OFPTYPE_ROLE_REPLY:
4243 case OFPTYPE_FLOW_MONITOR_PAUSED:
4244 case OFPTYPE_FLOW_MONITOR_RESUMED:
4245 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4246 case OFPTYPE_GET_ASYNC_REPLY:
4247 case OFPTYPE_GROUP_REPLY:
4248 case OFPTYPE_GROUP_DESC_REPLY:
4249 case OFPTYPE_GROUP_FEATURES_REPLY:
4250 case OFPTYPE_METER_REPLY:
4251 case OFPTYPE_METER_CONFIG_REPLY:
4252 case OFPTYPE_METER_FEATURES_REPLY:
4253 case OFPTYPE_TABLE_FEATURES_REPLY:
4255 return OFPERR_OFPBRC_BAD_TYPE;
4260 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4262 int error = handle_openflow__(ofconn, ofp_msg);
4263 if (error && error != OFPROTO_POSTPONE) {
4264 ofconn_send_error(ofconn, ofp_msg->data, error);
4266 COVERAGE_INC(ofproto_recv_openflow);
4267 return error != OFPROTO_POSTPONE;
4270 /* Asynchronous operations. */
4272 /* Creates and returns a new ofopgroup that is not associated with any
4273 * OpenFlow connection.
4275 * The caller should add operations to the returned group with
4276 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4277 static struct ofopgroup *
4278 ofopgroup_create_unattached(struct ofproto *ofproto)
4280 struct ofopgroup *group = xzalloc(sizeof *group);
4281 group->ofproto = ofproto;
4282 list_init(&group->ofproto_node);
4283 list_init(&group->ops);
4284 list_init(&group->ofconn_node);
4288 /* Creates and returns a new ofopgroup for 'ofproto'.
4290 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4291 * connection. The 'request' and 'buffer_id' arguments are ignored.
4293 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4294 * If the ofopgroup eventually fails, then the error reply will include
4295 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4296 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4298 * The caller should add operations to the returned group with
4299 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4300 static struct ofopgroup *
4301 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4302 const struct ofp_header *request, uint32_t buffer_id)
4304 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4306 size_t request_len = ntohs(request->length);
4308 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4310 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4311 group->ofconn = ofconn;
4312 group->request = xmemdup(request, MIN(request_len, 64));
4313 group->buffer_id = buffer_id;
4318 /* Submits 'group' for processing.
4320 * If 'group' contains no operations (e.g. none were ever added, or all of the
4321 * ones that were added completed synchronously), then it is destroyed
4322 * immediately. Otherwise it is added to the ofproto's list of pending
4325 ofopgroup_submit(struct ofopgroup *group)
4327 if (!group->n_running) {
4328 ofopgroup_complete(group);
4330 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4331 group->ofproto->n_pending++;
4336 ofopgroup_complete(struct ofopgroup *group)
4338 struct ofproto *ofproto = group->ofproto;
4340 struct ofconn *abbrev_ofconn;
4341 ovs_be32 abbrev_xid;
4343 struct ofoperation *op, *next_op;
4346 ovs_assert(!group->n_running);
4349 LIST_FOR_EACH (op, group_node, &group->ops) {
4356 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4357 LIST_FOR_EACH (op, group_node, &group->ops) {
4358 if (op->type != OFOPERATION_DELETE) {
4359 struct ofpbuf *packet;
4362 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4366 error = rule_execute(op->rule, in_port, packet);
4373 if (!error && !list_is_empty(&group->ofconn_node)) {
4374 abbrev_ofconn = group->ofconn;
4375 abbrev_xid = group->request->xid;
4377 abbrev_ofconn = NULL;
4378 abbrev_xid = htonl(0);
4380 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4381 struct rule *rule = op->rule;
4383 /* We generally want to report the change to active OpenFlow flow
4384 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4386 - The operation failed.
4388 - The affected rule is not visible to controllers.
4390 - The operation's only effect was to update rule->modified. */
4392 || ofproto_rule_is_hidden(rule)
4393 || (op->type == OFOPERATION_MODIFY
4395 && rule->flow_cookie == op->flow_cookie))) {
4396 /* Check that we can just cast from ofoperation_type to
4397 * nx_flow_update_event. */
4398 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4400 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4402 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4405 ofmonitor_report(ofproto->connmgr, rule,
4406 (enum nx_flow_update_event) op->type,
4407 op->reason, abbrev_ofconn, abbrev_xid);
4410 rule->pending = NULL;
4413 case OFOPERATION_ADD:
4417 ofproto_rule_destroy__(op->victim);
4418 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4419 if (vid_mask == VLAN_VID_MASK) {
4420 if (ofproto->vlan_bitmap) {
4421 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4422 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4423 bitmap_set1(ofproto->vlan_bitmap, vid);
4424 ofproto->vlans_changed = true;
4427 ofproto->vlans_changed = true;
4431 oftable_substitute_rule(rule, op->victim);
4432 ofproto_rule_destroy__(rule);
4436 case OFOPERATION_DELETE:
4437 ovs_assert(!op->error);
4438 ofproto_rule_destroy__(rule);
4442 case OFOPERATION_MODIFY:
4444 rule->modified = time_msec();
4446 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
4448 free(rule->ofpacts);
4449 rule->ofpacts = op->ofpacts;
4450 rule->ofpacts_len = op->ofpacts_len;
4452 op->ofpacts_len = 0;
4461 ofoperation_destroy(op);
4464 ofmonitor_flush(ofproto->connmgr);
4466 if (!list_is_empty(&group->ofproto_node)) {
4467 ovs_assert(ofproto->n_pending > 0);
4468 ofproto->n_pending--;
4469 list_remove(&group->ofproto_node);
4471 if (!list_is_empty(&group->ofconn_node)) {
4472 list_remove(&group->ofconn_node);
4474 ofconn_send_error(group->ofconn, group->request, error);
4476 connmgr_retry(ofproto->connmgr);
4478 free(group->request);
4482 /* Initiates a new operation on 'rule', of the specified 'type', within
4483 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4485 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4486 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4488 * Returns the newly created ofoperation (which is also available as
4489 * rule->pending). */
4490 static struct ofoperation *
4491 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4492 enum ofoperation_type type,
4493 enum ofp_flow_removed_reason reason)
4495 struct ofproto *ofproto = group->ofproto;
4496 struct ofoperation *op;
4498 ovs_assert(!rule->pending);
4500 op = rule->pending = xzalloc(sizeof *op);
4502 list_push_back(&group->ops, &op->group_node);
4505 op->reason = reason;
4506 op->flow_cookie = rule->flow_cookie;
4510 if (type == OFOPERATION_DELETE) {
4511 hmap_insert(&ofproto->deletions, &op->hmap_node,
4512 cls_rule_hash(&rule->cr, rule->table_id));
4519 ofoperation_destroy(struct ofoperation *op)
4521 struct ofopgroup *group = op->group;
4524 op->rule->pending = NULL;
4526 if (op->type == OFOPERATION_DELETE) {
4527 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4529 list_remove(&op->group_node);
4534 /* Indicates that 'op' completed with status 'error', which is either 0 to
4535 * indicate success or an OpenFlow error code on failure.
4537 * If 'error' is 0, indicating success, the operation will be committed
4538 * permanently to the flow table. There is one interesting subcase:
4540 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4541 * the flow table (the "victim" rule) by a new one, then the caller must
4542 * have uninitialized any derived state in the victim rule, as in step 5 in
4543 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4544 * performs steps 6 and 7 for the victim rule, most notably by calling its
4545 * ->rule_dealloc() function.
4547 * If 'error' is nonzero, then generally the operation will be rolled back:
4549 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4550 * restores the original rule. The caller must have uninitialized any
4551 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4552 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4553 * and 7 for the new rule, calling its ->rule_dealloc() function.
4555 * - If 'op' is a "modify flow" operation, ofproto restores the original
4558 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4559 * allowed to fail. It must always succeed.
4561 * Please see the large comment in ofproto/ofproto-provider.h titled
4562 * "Asynchronous Operation Support" for more information. */
4564 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4566 struct ofopgroup *group = op->group;
4568 ovs_assert(op->rule->pending == op);
4569 ovs_assert(group->n_running > 0);
4570 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4573 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4574 ofopgroup_complete(group);
4579 ofoperation_get_victim(struct ofoperation *op)
4581 ovs_assert(op->type == OFOPERATION_ADD);
4586 pick_datapath_id(const struct ofproto *ofproto)
4588 const struct ofport *port;
4590 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4592 uint8_t ea[ETH_ADDR_LEN];
4595 error = netdev_get_etheraddr(port->netdev, ea);
4597 return eth_addr_to_uint64(ea);
4599 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4600 ofproto->name, netdev_get_name(port->netdev),
4603 return ofproto->fallback_dpid;
4607 pick_fallback_dpid(void)
4609 uint8_t ea[ETH_ADDR_LEN];
4610 eth_addr_nicira_random(ea);
4611 return eth_addr_to_uint64(ea);
4614 /* Table overflow policy. */
4616 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4617 * is not configured to evict rules or if the table contains no evictable
4618 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4620 static struct rule *
4621 choose_rule_to_evict(struct oftable *table)
4623 struct eviction_group *evg;
4625 if (!table->eviction_fields) {
4629 /* In the common case, the outer and inner loops here will each be entered
4632 * - The inner loop normally "return"s in its first iteration. If the
4633 * eviction group has any evictable rules, then it always returns in
4636 * - The outer loop only iterates more than once if the largest eviction
4637 * group has no evictable rules.
4639 * - The outer loop can exit only if table's 'max_flows' is all filled up
4640 * by unevictable rules'. */
4641 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4644 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4645 if (rule->evictable) {
4654 /* Searches 'ofproto' for tables that have more flows than their configured
4655 * maximum and that have flow eviction enabled, and evicts as many flows as
4656 * necessary and currently feasible from them.
4658 * This triggers only when an OpenFlow table has N flows in it and then the
4659 * client configures a maximum number of flows less than N. */
4661 ofproto_evict(struct ofproto *ofproto)
4663 struct ofopgroup *group;
4664 struct oftable *table;
4666 group = ofopgroup_create_unattached(ofproto);
4667 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4668 while (classifier_count(&table->cls) > table->max_flows
4669 && table->eviction_fields) {
4672 rule = choose_rule_to_evict(table);
4673 if (!rule || rule->pending) {
4677 ofoperation_create(group, rule,
4678 OFOPERATION_DELETE, OFPRR_EVICTION);
4679 oftable_remove_rule(rule);
4680 ofproto->ofproto_class->rule_destruct(rule);
4683 ofopgroup_submit(group);
4686 /* Eviction groups. */
4688 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4689 * rules. The priority contains low-order random bits to ensure that eviction
4690 * groups with the same number of rules are prioritized randomly. */
4692 eviction_group_priority(size_t n_rules)
4694 uint16_t size = MIN(UINT16_MAX, n_rules);
4695 return (size << 16) | random_uint16();
4698 /* Updates 'evg', an eviction_group within 'table', following a change that
4699 * adds or removes rules in 'evg'. */
4701 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4703 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4704 eviction_group_priority(heap_count(&evg->rules)));
4707 /* Destroys 'evg', an eviction_group within 'table':
4709 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4710 * rules themselves, just removes them from the eviction group.)
4712 * - Removes 'evg' from 'table'.
4716 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4718 while (!heap_is_empty(&evg->rules)) {
4721 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4722 rule->eviction_group = NULL;
4724 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4725 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4726 heap_destroy(&evg->rules);
4730 /* Removes 'rule' from its eviction group, if any. */
4732 eviction_group_remove_rule(struct rule *rule)
4734 if (rule->eviction_group) {
4735 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4736 struct eviction_group *evg = rule->eviction_group;
4738 rule->eviction_group = NULL;
4739 heap_remove(&evg->rules, &rule->evg_node);
4740 if (heap_is_empty(&evg->rules)) {
4741 eviction_group_destroy(table, evg);
4743 eviction_group_resized(table, evg);
4748 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4749 * returns the hash value. */
4751 eviction_group_hash_rule(struct rule *rule)
4753 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4754 const struct mf_subfield *sf;
4758 hash = table->eviction_group_id_basis;
4759 miniflow_expand(&rule->cr.match.flow, &flow);
4760 for (sf = table->eviction_fields;
4761 sf < &table->eviction_fields[table->n_eviction_fields];
4764 if (mf_are_prereqs_ok(sf->field, &flow)) {
4765 union mf_value value;
4767 mf_get_value(sf->field, &flow, &value);
4769 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4771 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4772 unsigned int start = sf->ofs + sf->n_bits;
4773 bitwise_zero(&value, sf->field->n_bytes, start,
4774 sf->field->n_bytes * 8 - start);
4776 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4778 hash = hash_int(hash, 0);
4785 /* Returns an eviction group within 'table' with the given 'id', creating one
4787 static struct eviction_group *
4788 eviction_group_find(struct oftable *table, uint32_t id)
4790 struct eviction_group *evg;
4792 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4796 evg = xmalloc(sizeof *evg);
4797 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4798 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4799 eviction_group_priority(0));
4800 heap_init(&evg->rules);
4805 /* Returns an eviction priority for 'rule'. The return value should be
4806 * interpreted so that higher priorities make a rule more attractive candidates
4809 rule_eviction_priority(struct rule *rule)
4811 long long int hard_expiration;
4812 long long int idle_expiration;
4813 long long int expiration;
4814 uint32_t expiration_offset;
4816 /* Calculate time of expiration. */
4817 hard_expiration = (rule->hard_timeout
4818 ? rule->modified + rule->hard_timeout * 1000
4820 idle_expiration = (rule->idle_timeout
4821 ? rule->used + rule->idle_timeout * 1000
4823 expiration = MIN(hard_expiration, idle_expiration);
4824 if (expiration == LLONG_MAX) {
4828 /* Calculate the time of expiration as a number of (approximate) seconds
4829 * after program startup.
4831 * This should work OK for program runs that last UINT32_MAX seconds or
4832 * less. Therefore, please restart OVS at least once every 136 years. */
4833 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4835 /* Invert the expiration offset because we're using a max-heap. */
4836 return UINT32_MAX - expiration_offset;
4839 /* Adds 'rule' to an appropriate eviction group for its oftable's
4840 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4841 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4844 * The caller must ensure that 'rule' is not already in an eviction group. */
4846 eviction_group_add_rule(struct rule *rule)
4848 struct ofproto *ofproto = rule->ofproto;
4849 struct oftable *table = &ofproto->tables[rule->table_id];
4851 if (table->eviction_fields
4852 && (rule->hard_timeout || rule->idle_timeout)) {
4853 struct eviction_group *evg;
4855 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4857 rule->eviction_group = evg;
4858 heap_insert(&evg->rules, &rule->evg_node,
4859 rule_eviction_priority(rule));
4860 eviction_group_resized(table, evg);
4866 /* Initializes 'table'. */
4868 oftable_init(struct oftable *table)
4870 memset(table, 0, sizeof *table);
4871 classifier_init(&table->cls);
4872 table->max_flows = UINT_MAX;
4875 /* Destroys 'table', including its classifier and eviction groups.
4877 * The caller is responsible for freeing 'table' itself. */
4879 oftable_destroy(struct oftable *table)
4881 ovs_assert(classifier_is_empty(&table->cls));
4882 oftable_disable_eviction(table);
4883 classifier_destroy(&table->cls);
4887 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4888 * string, then 'table' will use its default name.
4890 * This only affects the name exposed for a table exposed through the OpenFlow
4891 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4893 oftable_set_name(struct oftable *table, const char *name)
4895 if (name && name[0]) {
4896 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4897 if (!table->name || strncmp(name, table->name, len)) {
4899 table->name = xmemdup0(name, len);
4907 /* oftables support a choice of two policies when adding a rule would cause the
4908 * number of flows in the table to exceed the configured maximum number: either
4909 * they can refuse to add the new flow or they can evict some existing flow.
4910 * This function configures the former policy on 'table'. */
4912 oftable_disable_eviction(struct oftable *table)
4914 if (table->eviction_fields) {
4915 struct eviction_group *evg, *next;
4917 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4918 &table->eviction_groups_by_id) {
4919 eviction_group_destroy(table, evg);
4921 hmap_destroy(&table->eviction_groups_by_id);
4922 heap_destroy(&table->eviction_groups_by_size);
4924 free(table->eviction_fields);
4925 table->eviction_fields = NULL;
4926 table->n_eviction_fields = 0;
4930 /* oftables support a choice of two policies when adding a rule would cause the
4931 * number of flows in the table to exceed the configured maximum number: either
4932 * they can refuse to add the new flow or they can evict some existing flow.
4933 * This function configures the latter policy on 'table', with fairness based
4934 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4935 * 'n_fields' as 0 disables fairness.) */
4937 oftable_enable_eviction(struct oftable *table,
4938 const struct mf_subfield *fields, size_t n_fields)
4940 struct cls_cursor cursor;
4943 if (table->eviction_fields
4944 && n_fields == table->n_eviction_fields
4946 || !memcmp(fields, table->eviction_fields,
4947 n_fields * sizeof *fields))) {
4952 oftable_disable_eviction(table);
4954 table->n_eviction_fields = n_fields;
4955 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4957 table->eviction_group_id_basis = random_uint32();
4958 hmap_init(&table->eviction_groups_by_id);
4959 heap_init(&table->eviction_groups_by_size);
4961 cls_cursor_init(&cursor, &table->cls, NULL);
4962 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4963 eviction_group_add_rule(rule);
4967 /* Removes 'rule' from the oftable that contains it. */
4969 oftable_remove_rule(struct rule *rule)
4971 struct ofproto *ofproto = rule->ofproto;
4972 struct oftable *table = &ofproto->tables[rule->table_id];
4974 classifier_remove(&table->cls, &rule->cr);
4975 cookies_remove(ofproto, rule);
4976 eviction_group_remove_rule(rule);
4977 if (!list_is_empty(&rule->expirable)) {
4978 list_remove(&rule->expirable);
4982 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4983 * oftable that has an identical cls_rule. Returns the rule that was removed,
4984 * if any, and otherwise NULL. */
4985 static struct rule *
4986 oftable_replace_rule(struct rule *rule)
4988 struct ofproto *ofproto = rule->ofproto;
4989 struct oftable *table = &ofproto->tables[rule->table_id];
4990 struct rule *victim;
4991 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4994 list_insert(&ofproto->expirable, &rule->expirable);
4996 cookies_insert(ofproto, rule);
4998 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
5000 cookies_remove(ofproto, victim);
5002 if (!list_is_empty(&victim->expirable)) {
5003 list_remove(&victim->expirable);
5005 eviction_group_remove_rule(victim);
5007 eviction_group_add_rule(rule);
5011 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
5013 oftable_substitute_rule(struct rule *old, struct rule *new)
5016 oftable_replace_rule(new);
5018 oftable_remove_rule(old);
5022 /* unixctl commands. */
5025 ofproto_lookup(const char *name)
5027 struct ofproto *ofproto;
5029 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5031 if (!strcmp(ofproto->name, name)) {
5039 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5040 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5042 struct ofproto *ofproto;
5046 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5047 ds_put_format(&results, "%s\n", ofproto->name);
5049 unixctl_command_reply(conn, ds_cstr(&results));
5050 ds_destroy(&results);
5054 ofproto_unixctl_init(void)
5056 static bool registered;
5062 unixctl_command_register("ofproto/list", "", 0, 0,
5063 ofproto_unixctl_list, NULL);
5066 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5068 * This is deprecated. It is only for compatibility with broken device drivers
5069 * in old versions of Linux that do not properly support VLANs when VLAN
5070 * devices are not used. When broken device drivers are no longer in
5071 * widespread use, we will delete these interfaces. */
5073 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5074 * (exactly) by an OpenFlow rule in 'ofproto'. */
5076 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5078 const struct oftable *oftable;
5080 free(ofproto->vlan_bitmap);
5081 ofproto->vlan_bitmap = bitmap_allocate(4096);
5082 ofproto->vlans_changed = false;
5084 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5085 const struct cls_table *table;
5087 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5088 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5089 const struct cls_rule *rule;
5091 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5092 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5093 bitmap_set1(vlan_bitmap, vid);
5094 bitmap_set1(ofproto->vlan_bitmap, vid);
5101 /* Returns true if new VLANs have come into use by the flow table since the
5102 * last call to ofproto_get_vlan_usage().
5104 * We don't track when old VLANs stop being used. */
5106 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5108 return ofproto->vlans_changed;
5111 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5112 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5113 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5114 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5115 * then the VLAN device is un-enslaved. */
5117 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5118 ofp_port_t realdev_ofp_port, int vid)
5120 struct ofport *ofport;
5123 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5125 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5127 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5128 ofproto->name, vlandev_ofp_port);
5132 if (!ofproto->ofproto_class->set_realdev) {
5133 if (!vlandev_ofp_port) {
5136 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5140 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5142 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5143 ofproto->name, vlandev_ofp_port,
5144 netdev_get_name(ofport->netdev), strerror(error));