2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
208 static uint64_t pick_datapath_id(const struct ofproto *);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto *);
211 static void update_mtu(struct ofproto *, struct ofport *);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class **ofproto_classes;
218 static size_t n_ofproto_classes;
219 static size_t allocated_ofproto_classes;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
224 /* Initial mappings of port to OpenFlow number mappings. */
225 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
229 /* The default value of true waits for flow restore. */
230 static bool flow_restore_wait = true;
232 /* Must be called to initialize the ofproto library.
234 * The caller may pass in 'iface_hints', which contains an shash of
235 * "iface_hint" elements indexed by the interface's name. The provider
236 * may use these hints to describe the startup configuration in order to
237 * reinitialize its state. The caller owns the provided data, so a
238 * provider will make copies of anything required. An ofproto provider
239 * will remove any existing state that is not described by the hint, and
240 * may choose to remove it all. */
242 ofproto_init(const struct shash *iface_hints)
244 struct shash_node *node;
247 ofproto_class_register(&ofproto_dpif_class);
249 /* Make a local copy, since we don't own 'iface_hints' elements. */
250 SHASH_FOR_EACH(node, iface_hints) {
251 const struct iface_hint *orig_hint = node->data;
252 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
253 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
255 new_hint->br_name = xstrdup(orig_hint->br_name);
256 new_hint->br_type = xstrdup(br_type);
257 new_hint->ofp_port = orig_hint->ofp_port;
259 shash_add(&init_ofp_ports, node->name, new_hint);
262 for (i = 0; i < n_ofproto_classes; i++) {
263 ofproto_classes[i]->init(&init_ofp_ports);
267 /* 'type' should be a normalized datapath type, as returned by
268 * ofproto_normalize_type(). Returns the corresponding ofproto_class
269 * structure, or a null pointer if there is none registered for 'type'. */
270 static const struct ofproto_class *
271 ofproto_class_find__(const char *type)
275 for (i = 0; i < n_ofproto_classes; i++) {
276 const struct ofproto_class *class = ofproto_classes[i];
281 class->enumerate_types(&types);
282 found = sset_contains(&types, type);
283 sset_destroy(&types);
289 VLOG_WARN("unknown datapath type %s", type);
293 /* Registers a new ofproto class. After successful registration, new ofprotos
294 * of that type can be created using ofproto_create(). */
296 ofproto_class_register(const struct ofproto_class *new_class)
300 for (i = 0; i < n_ofproto_classes; i++) {
301 if (ofproto_classes[i] == new_class) {
306 if (n_ofproto_classes >= allocated_ofproto_classes) {
307 ofproto_classes = x2nrealloc(ofproto_classes,
308 &allocated_ofproto_classes,
309 sizeof *ofproto_classes);
311 ofproto_classes[n_ofproto_classes++] = new_class;
315 /* Unregisters a datapath provider. 'type' must have been previously
316 * registered and not currently be in use by any ofprotos. After
317 * unregistration new datapaths of that type cannot be opened using
318 * ofproto_create(). */
320 ofproto_class_unregister(const struct ofproto_class *class)
324 for (i = 0; i < n_ofproto_classes; i++) {
325 if (ofproto_classes[i] == class) {
326 for (i++; i < n_ofproto_classes; i++) {
327 ofproto_classes[i - 1] = ofproto_classes[i];
333 VLOG_WARN("attempted to unregister an ofproto class that is not "
338 /* Clears 'types' and enumerates all registered ofproto types into it. The
339 * caller must first initialize the sset. */
341 ofproto_enumerate_types(struct sset *types)
345 for (i = 0; i < n_ofproto_classes; i++) {
346 ofproto_classes[i]->enumerate_types(types);
350 /* Returns the fully spelled out name for the given ofproto 'type'.
352 * Normalized type string can be compared with strcmp(). Unnormalized type
353 * string might be the same even if they have different spellings. */
355 ofproto_normalize_type(const char *type)
357 return type && type[0] ? type : "system";
360 /* Clears 'names' and enumerates the names of all known created ofprotos with
361 * the given 'type'. The caller must first initialize the sset. Returns 0 if
362 * successful, otherwise a positive errno value.
364 * Some kinds of datapaths might not be practically enumerable. This is not
365 * considered an error. */
367 ofproto_enumerate_names(const char *type, struct sset *names)
369 const struct ofproto_class *class = ofproto_class_find__(type);
370 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
374 ofproto_create(const char *datapath_name, const char *datapath_type,
375 struct ofproto **ofprotop)
377 const struct ofproto_class *class;
378 struct ofproto *ofproto;
384 ofproto_unixctl_init();
386 datapath_type = ofproto_normalize_type(datapath_type);
387 class = ofproto_class_find__(datapath_type);
389 VLOG_WARN("could not create datapath %s of unknown type %s",
390 datapath_name, datapath_type);
394 ofproto = class->alloc();
396 VLOG_ERR("failed to allocate datapath %s of type %s",
397 datapath_name, datapath_type);
402 memset(ofproto, 0, sizeof *ofproto);
403 ofproto->ofproto_class = class;
404 ofproto->name = xstrdup(datapath_name);
405 ofproto->type = xstrdup(datapath_type);
406 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
407 hash_string(ofproto->name, 0));
408 ofproto->datapath_id = 0;
409 ofproto_set_flow_eviction_threshold(ofproto,
410 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
411 ofproto->forward_bpdu = false;
412 ofproto->fallback_dpid = pick_fallback_dpid();
413 ofproto->mfr_desc = NULL;
414 ofproto->hw_desc = NULL;
415 ofproto->sw_desc = NULL;
416 ofproto->serial_desc = NULL;
417 ofproto->dp_desc = NULL;
418 ofproto->frag_handling = OFPC_FRAG_NORMAL;
419 hmap_init(&ofproto->ports);
420 shash_init(&ofproto->port_by_name);
421 simap_init(&ofproto->ofp_requests);
422 ofproto->max_ports = OFPP_MAX;
423 ofproto->tables = NULL;
424 ofproto->n_tables = 0;
425 list_init(&ofproto->expirable);
426 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
427 ofproto->state = S_OPENFLOW;
428 list_init(&ofproto->pending);
429 ofproto->n_pending = 0;
430 hmap_init(&ofproto->deletions);
431 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
432 ofproto->first_op = ofproto->last_op = LLONG_MIN;
433 ofproto->next_op_report = LLONG_MAX;
434 ofproto->op_backoff = LLONG_MIN;
435 ofproto->vlan_bitmap = NULL;
436 ofproto->vlans_changed = false;
437 ofproto->min_mtu = INT_MAX;
439 error = ofproto->ofproto_class->construct(ofproto);
441 VLOG_ERR("failed to open datapath %s: %s",
442 datapath_name, strerror(error));
443 ofproto_destroy__(ofproto);
447 /* The "max_ports" member should have been set by ->construct(ofproto).
448 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
449 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
450 bitmap_set1(ofproto->ofp_port_ids, 0);
452 /* Check that hidden tables, if any, are at the end. */
453 ovs_assert(ofproto->n_tables);
454 for (i = 0; i + 1 < ofproto->n_tables; i++) {
455 enum oftable_flags flags = ofproto->tables[i].flags;
456 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
458 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
461 ofproto->datapath_id = pick_datapath_id(ofproto);
468 /* Must be called (only) by an ofproto implementation in its constructor
469 * function. See the large comment on 'construct' in struct ofproto_class for
472 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
474 struct oftable *table;
476 ovs_assert(!ofproto->n_tables);
477 ovs_assert(n_tables >= 1 && n_tables <= 255);
479 ofproto->n_tables = n_tables;
480 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
481 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
486 /* To be optionally called (only) by an ofproto implementation in its
487 * constructor function. See the large comment on 'construct' in struct
488 * ofproto_class for details.
490 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
491 * will then ensure that actions passed into the ofproto implementation will
492 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
493 * function is not called, there will be no such restriction.
495 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
496 * the 'max_ports' restriction. */
498 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
500 ovs_assert(max_ports <= OFPP_MAX);
501 ofproto->max_ports = max_ports;
505 ofproto_get_datapath_id(const struct ofproto *ofproto)
507 return ofproto->datapath_id;
511 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
513 uint64_t old_dpid = p->datapath_id;
514 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
515 if (p->datapath_id != old_dpid) {
516 /* Force all active connections to reconnect, since there is no way to
517 * notify a controller that the datapath ID has changed. */
518 ofproto_reconnect_controllers(p);
523 ofproto_set_controllers(struct ofproto *p,
524 const struct ofproto_controller *controllers,
525 size_t n_controllers, uint32_t allowed_versions)
527 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
532 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
534 connmgr_set_fail_mode(p->connmgr, fail_mode);
537 /* Drops the connections between 'ofproto' and all of its controllers, forcing
538 * them to reconnect. */
540 ofproto_reconnect_controllers(struct ofproto *ofproto)
542 connmgr_reconnect(ofproto->connmgr);
545 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
546 * in-band control should guarantee access, in the same way that in-band
547 * control guarantees access to OpenFlow controllers. */
549 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
550 const struct sockaddr_in *extras, size_t n)
552 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
555 /* Sets the OpenFlow queue used by flows set up by in-band control on
556 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
557 * flows will use the default queue. */
559 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
561 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
564 /* Sets the number of flows at which eviction from the kernel flow table
567 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
569 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
570 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
572 ofproto->flow_eviction_threshold = threshold;
576 /* If forward_bpdu is true, the NORMAL action will forward frames with
577 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
578 * the NORMAL action will drop these frames. */
580 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
582 bool old_val = ofproto->forward_bpdu;
583 ofproto->forward_bpdu = forward_bpdu;
584 if (old_val != ofproto->forward_bpdu) {
585 if (ofproto->ofproto_class->forward_bpdu_changed) {
586 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
591 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
592 * 'idle_time', in seconds, and the maximum number of MAC table entries to
595 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
598 if (ofproto->ofproto_class->set_mac_table_config) {
599 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
605 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
608 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
612 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
614 return connmgr_set_snoops(ofproto->connmgr, snoops);
618 ofproto_set_netflow(struct ofproto *ofproto,
619 const struct netflow_options *nf_options)
621 if (nf_options && sset_is_empty(&nf_options->collectors)) {
625 if (ofproto->ofproto_class->set_netflow) {
626 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
628 return nf_options ? EOPNOTSUPP : 0;
633 ofproto_set_sflow(struct ofproto *ofproto,
634 const struct ofproto_sflow_options *oso)
636 if (oso && sset_is_empty(&oso->targets)) {
640 if (ofproto->ofproto_class->set_sflow) {
641 return ofproto->ofproto_class->set_sflow(ofproto, oso);
643 return oso ? EOPNOTSUPP : 0;
648 ofproto_set_ipfix(struct ofproto *ofproto,
649 const struct ofproto_ipfix_bridge_exporter_options *bo,
650 const struct ofproto_ipfix_flow_exporter_options *fo,
653 if (ofproto->ofproto_class->set_ipfix) {
654 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
656 return (bo || fo) ? EOPNOTSUPP : 0;
661 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
663 flow_restore_wait = flow_restore_wait_db;
667 ofproto_get_flow_restore_wait(void)
669 return flow_restore_wait;
673 /* Spanning Tree Protocol (STP) configuration. */
675 /* Configures STP on 'ofproto' using the settings defined in 's'. If
676 * 's' is NULL, disables STP.
678 * Returns 0 if successful, otherwise a positive errno value. */
680 ofproto_set_stp(struct ofproto *ofproto,
681 const struct ofproto_stp_settings *s)
683 return (ofproto->ofproto_class->set_stp
684 ? ofproto->ofproto_class->set_stp(ofproto, s)
688 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
689 * 'enabled' member of 's' is false, then the other members are not
692 * Returns 0 if successful, otherwise a positive errno value. */
694 ofproto_get_stp_status(struct ofproto *ofproto,
695 struct ofproto_stp_status *s)
697 return (ofproto->ofproto_class->get_stp_status
698 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
702 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
703 * in 's'. The caller is responsible for assigning STP port numbers
704 * (using the 'port_num' member in the range of 1 through 255, inclusive)
705 * and ensuring there are no duplicates. If the 's' is NULL, then STP
706 * is disabled on the port.
708 * Returns 0 if successful, otherwise a positive errno value.*/
710 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
711 const struct ofproto_port_stp_settings *s)
713 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
715 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
716 ofproto->name, ofp_port);
720 return (ofproto->ofproto_class->set_stp_port
721 ? ofproto->ofproto_class->set_stp_port(ofport, s)
725 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
726 * 's'. If the 'enabled' member in 's' is false, then the other members
727 * are not meaningful.
729 * Returns 0 if successful, otherwise a positive errno value.*/
731 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
732 struct ofproto_port_stp_status *s)
734 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
736 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
737 "port %"PRIu16, ofproto->name, ofp_port);
741 return (ofproto->ofproto_class->get_stp_port_status
742 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
746 /* Queue DSCP configuration. */
748 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
749 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
750 * to implement QoS. Instead, it is used to implement features which require
751 * knowledge of what queues exist on a port, and some basic information about
754 * Returns 0 if successful, otherwise a positive errno value. */
756 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
757 const struct ofproto_port_queue *queues,
760 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
763 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
764 ofproto->name, ofp_port);
768 return (ofproto->ofproto_class->set_queues
769 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
773 /* Connectivity Fault Management configuration. */
775 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
777 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
779 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
780 if (ofport && ofproto->ofproto_class->set_cfm) {
781 ofproto->ofproto_class->set_cfm(ofport, NULL);
785 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
786 * basic configuration from the configuration members in 'cfm', and the remote
787 * maintenance point ID from remote_mpid. Ignores the statistics members of
790 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
792 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
793 const struct cfm_settings *s)
795 struct ofport *ofport;
798 ofport = ofproto_get_port(ofproto, ofp_port);
800 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
801 ofproto->name, ofp_port);
805 /* XXX: For configuration simplicity, we only support one remote_mpid
806 * outside of the CFM module. It's not clear if this is the correct long
807 * term solution or not. */
808 error = (ofproto->ofproto_class->set_cfm
809 ? ofproto->ofproto_class->set_cfm(ofport, s)
812 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
813 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
818 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
819 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
820 * 0 if LACP partner information is not current (generally indicating a
821 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
823 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
825 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
826 return (ofport && ofproto->ofproto_class->port_is_lacp_current
827 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
833 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
834 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
835 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
836 * configuration plus, if there is more than one slave, a bonding
839 * If 'aux' is already registered then this function updates its configuration
840 * to 's'. Otherwise, this function registers a new bundle.
842 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
845 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
846 const struct ofproto_bundle_settings *s)
848 return (ofproto->ofproto_class->bundle_set
849 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
853 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
854 * If no such bundle has been registered, this has no effect. */
856 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
858 return ofproto_bundle_register(ofproto, aux, NULL);
862 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
863 * If 'aux' is already registered then this function updates its configuration
864 * to 's'. Otherwise, this function registers a new mirror. */
866 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
867 const struct ofproto_mirror_settings *s)
869 return (ofproto->ofproto_class->mirror_set
870 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
874 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
875 * If no mirror has been registered, this has no effect. */
877 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
879 return ofproto_mirror_register(ofproto, aux, NULL);
882 /* Retrieves statistics from mirror associated with client data pointer
883 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
884 * 'bytes', respectively. If a particular counters is not supported,
885 * the appropriate argument is set to UINT64_MAX. */
887 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
888 uint64_t *packets, uint64_t *bytes)
890 if (!ofproto->ofproto_class->mirror_get_stats) {
891 *packets = *bytes = UINT64_MAX;
895 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
899 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
900 * which all packets are flooded, instead of using MAC learning. If
901 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
903 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
906 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
908 return (ofproto->ofproto_class->set_flood_vlans
909 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
913 /* Returns true if 'aux' is a registered bundle that is currently in use as the
914 * output for a mirror. */
916 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
918 return (ofproto->ofproto_class->is_mirror_output_bundle
919 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
923 /* Configuration of OpenFlow tables. */
925 /* Returns the number of OpenFlow tables in 'ofproto'. */
927 ofproto_get_n_tables(const struct ofproto *ofproto)
929 return ofproto->n_tables;
932 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
933 * settings from 's'. 'table_id' must be in the range 0 through the number of
934 * OpenFlow tables in 'ofproto' minus 1, inclusive.
936 * For read-only tables, only the name may be configured. */
938 ofproto_configure_table(struct ofproto *ofproto, int table_id,
939 const struct ofproto_table_settings *s)
941 struct oftable *table;
943 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
944 table = &ofproto->tables[table_id];
946 oftable_set_name(table, s->name);
948 if (table->flags & OFTABLE_READONLY) {
953 oftable_enable_eviction(table, s->groups, s->n_groups);
955 oftable_disable_eviction(table);
958 table->max_flows = s->max_flows;
959 if (classifier_count(&table->cls) > table->max_flows
960 && table->eviction_fields) {
961 /* 'table' contains more flows than allowed. We might not be able to
962 * evict them right away because of the asynchronous nature of flow
963 * table changes. Schedule eviction for later. */
964 switch (ofproto->state) {
966 ofproto->state = S_EVICT;
970 /* We're already deleting flows, nothing more to do. */
977 ofproto_has_snoops(const struct ofproto *ofproto)
979 return connmgr_has_snoops(ofproto->connmgr);
983 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
985 connmgr_get_snoops(ofproto->connmgr, snoops);
989 ofproto_flush__(struct ofproto *ofproto)
991 struct ofopgroup *group;
992 struct oftable *table;
994 if (ofproto->ofproto_class->flush) {
995 ofproto->ofproto_class->flush(ofproto);
998 group = ofopgroup_create_unattached(ofproto);
999 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1000 struct rule *rule, *next_rule;
1001 struct cls_cursor cursor;
1003 if (table->flags & OFTABLE_HIDDEN) {
1007 cls_cursor_init(&cursor, &table->cls, NULL);
1008 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1009 if (!rule->pending) {
1010 ofoperation_create(group, rule, OFOPERATION_DELETE,
1012 oftable_remove_rule(rule);
1013 ofproto->ofproto_class->rule_destruct(rule);
1017 ofopgroup_submit(group);
1021 ofproto_destroy__(struct ofproto *ofproto)
1023 struct oftable *table;
1025 ovs_assert(list_is_empty(&ofproto->pending));
1026 ovs_assert(!ofproto->n_pending);
1028 connmgr_destroy(ofproto->connmgr);
1030 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1031 free(ofproto->name);
1032 free(ofproto->type);
1033 free(ofproto->mfr_desc);
1034 free(ofproto->hw_desc);
1035 free(ofproto->sw_desc);
1036 free(ofproto->serial_desc);
1037 free(ofproto->dp_desc);
1038 hmap_destroy(&ofproto->ports);
1039 shash_destroy(&ofproto->port_by_name);
1040 bitmap_free(ofproto->ofp_port_ids);
1041 simap_destroy(&ofproto->ofp_requests);
1043 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1044 oftable_destroy(table);
1046 free(ofproto->tables);
1048 hmap_destroy(&ofproto->deletions);
1050 free(ofproto->vlan_bitmap);
1052 ofproto->ofproto_class->dealloc(ofproto);
1056 ofproto_destroy(struct ofproto *p)
1058 struct ofport *ofport, *next_ofport;
1065 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1066 ofport_destroy(ofport);
1069 p->ofproto_class->destruct(p);
1070 ofproto_destroy__(p);
1073 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1074 * kernel datapath, for example, this destroys the datapath in the kernel, and
1075 * with the netdev-based datapath, it tears down the data structures that
1076 * represent the datapath.
1078 * The datapath should not be currently open as an ofproto. */
1080 ofproto_delete(const char *name, const char *type)
1082 const struct ofproto_class *class = ofproto_class_find__(type);
1083 return (!class ? EAFNOSUPPORT
1084 : !class->del ? EACCES
1085 : class->del(type, name));
1089 process_port_change(struct ofproto *ofproto, int error, char *devname)
1091 if (error == ENOBUFS) {
1092 reinit_ports(ofproto);
1093 } else if (!error) {
1094 update_port(ofproto, devname);
1100 ofproto_type_run(const char *datapath_type)
1102 const struct ofproto_class *class;
1105 datapath_type = ofproto_normalize_type(datapath_type);
1106 class = ofproto_class_find__(datapath_type);
1108 error = class->type_run ? class->type_run(datapath_type) : 0;
1109 if (error && error != EAGAIN) {
1110 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1111 datapath_type, strerror(error));
1117 ofproto_type_run_fast(const char *datapath_type)
1119 const struct ofproto_class *class;
1122 datapath_type = ofproto_normalize_type(datapath_type);
1123 class = ofproto_class_find__(datapath_type);
1125 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1126 if (error && error != EAGAIN) {
1127 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1128 datapath_type, strerror(error));
1134 ofproto_type_wait(const char *datapath_type)
1136 const struct ofproto_class *class;
1138 datapath_type = ofproto_normalize_type(datapath_type);
1139 class = ofproto_class_find__(datapath_type);
1141 if (class->type_wait) {
1142 class->type_wait(datapath_type);
1147 ofproto_run(struct ofproto *p)
1149 struct sset changed_netdevs;
1150 const char *changed_netdev;
1151 struct ofport *ofport;
1154 error = p->ofproto_class->run(p);
1155 if (error && error != EAGAIN) {
1156 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1159 if (p->ofproto_class->port_poll) {
1162 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1163 process_port_change(p, error, devname);
1167 /* Update OpenFlow port status for any port whose netdev has changed.
1169 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1170 * destroyed, so it's not safe to update ports directly from the
1171 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1172 * need this two-phase approach. */
1173 sset_init(&changed_netdevs);
1174 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1175 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1176 if (ofport->change_seq != change_seq) {
1177 ofport->change_seq = change_seq;
1178 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1181 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1182 update_port(p, changed_netdev);
1184 sset_destroy(&changed_netdevs);
1188 connmgr_run(p->connmgr, handle_openflow);
1192 connmgr_run(p->connmgr, NULL);
1194 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1195 p->state = S_OPENFLOW;
1200 connmgr_run(p->connmgr, NULL);
1202 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1203 connmgr_flushed(p->connmgr);
1204 p->state = S_OPENFLOW;
1212 if (time_msec() >= p->next_op_report) {
1213 long long int ago = (time_msec() - p->first_op) / 1000;
1214 long long int interval = (p->last_op - p->first_op) / 1000;
1218 ds_put_format(&s, "%d flow_mods ",
1219 p->n_add + p->n_delete + p->n_modify);
1220 if (interval == ago) {
1221 ds_put_format(&s, "in the last %lld s", ago);
1222 } else if (interval) {
1223 ds_put_format(&s, "in the %lld s starting %lld s ago",
1226 ds_put_format(&s, "%lld s ago", ago);
1229 ds_put_cstr(&s, " (");
1231 ds_put_format(&s, "%d adds, ", p->n_add);
1234 ds_put_format(&s, "%d deletes, ", p->n_delete);
1237 ds_put_format(&s, "%d modifications, ", p->n_modify);
1240 ds_put_char(&s, ')');
1242 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1245 p->n_add = p->n_delete = p->n_modify = 0;
1246 p->next_op_report = LLONG_MAX;
1252 /* Performs periodic activity required by 'ofproto' that needs to be done
1253 * with the least possible latency.
1255 * It makes sense to call this function a couple of times per poll loop, to
1256 * provide a significant performance boost on some benchmarks with the
1257 * ofproto-dpif implementation. */
1259 ofproto_run_fast(struct ofproto *p)
1263 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1264 if (error && error != EAGAIN) {
1265 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1266 p->name, strerror(error));
1272 ofproto_wait(struct ofproto *p)
1274 struct ofport *ofport;
1276 p->ofproto_class->wait(p);
1277 if (p->ofproto_class->port_poll_wait) {
1278 p->ofproto_class->port_poll_wait(p);
1281 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1282 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1283 poll_immediate_wake();
1289 connmgr_wait(p->connmgr, true);
1294 connmgr_wait(p->connmgr, false);
1295 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1296 poll_immediate_wake();
1303 ofproto_is_alive(const struct ofproto *p)
1305 return connmgr_has_controllers(p->connmgr);
1308 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1309 * memory_report(). */
1311 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1313 const struct oftable *table;
1314 unsigned int n_rules;
1316 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1317 simap_increase(usage, "ops",
1318 ofproto->n_pending + hmap_count(&ofproto->deletions));
1321 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1322 n_rules += classifier_count(&table->cls);
1324 simap_increase(usage, "rules", n_rules);
1326 if (ofproto->ofproto_class->get_memory_usage) {
1327 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1330 connmgr_get_memory_usage(ofproto->connmgr, usage);
1334 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1337 connmgr_get_controller_info(ofproto->connmgr, info);
1341 ofproto_free_ofproto_controller_info(struct shash *info)
1343 connmgr_free_controller_info(info);
1346 /* Makes a deep copy of 'old' into 'port'. */
1348 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1350 port->name = xstrdup(old->name);
1351 port->type = xstrdup(old->type);
1352 port->ofp_port = old->ofp_port;
1355 /* Frees memory allocated to members of 'ofproto_port'.
1357 * Do not call this function on an ofproto_port obtained from
1358 * ofproto_port_dump_next(): that function retains ownership of the data in the
1361 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1363 free(ofproto_port->name);
1364 free(ofproto_port->type);
1367 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1369 * This function provides no status indication. An error status for the entire
1370 * dump operation is provided when it is completed by calling
1371 * ofproto_port_dump_done().
1374 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1375 const struct ofproto *ofproto)
1377 dump->ofproto = ofproto;
1378 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1382 /* Attempts to retrieve another port from 'dump', which must have been created
1383 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1384 * 'port' and returns true. On failure, returns false.
1386 * Failure might indicate an actual error or merely that the last port has been
1387 * dumped. An error status for the entire dump operation is provided when it
1388 * is completed by calling ofproto_port_dump_done().
1390 * The ofproto owns the data stored in 'port'. It will remain valid until at
1391 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1392 * ofproto_port_dump_done(). */
1394 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1395 struct ofproto_port *port)
1397 const struct ofproto *ofproto = dump->ofproto;
1403 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1406 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1412 /* Completes port table dump operation 'dump', which must have been created
1413 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1414 * error-free, otherwise a positive errno value describing the problem. */
1416 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1418 const struct ofproto *ofproto = dump->ofproto;
1420 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1423 return dump->error == EOF ? 0 : dump->error;
1426 /* Returns the type to pass to netdev_open() when a datapath of type
1427 * 'datapath_type' has a port of type 'port_type', for a few special
1428 * cases when a netdev type differs from a port type. For example, when
1429 * using the userspace datapath, a port of type "internal" needs to be
1432 * Returns either 'type' itself or a string literal, which must not be
1435 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1437 const struct ofproto_class *class;
1439 datapath_type = ofproto_normalize_type(datapath_type);
1440 class = ofproto_class_find__(datapath_type);
1445 return (class->port_open_type
1446 ? class->port_open_type(datapath_type, port_type)
1450 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1451 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1452 * the port's OpenFlow port number.
1454 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1455 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1456 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1457 * 'ofp_portp' is non-null). */
1459 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1460 uint16_t *ofp_portp)
1462 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1465 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1467 const char *netdev_name = netdev_get_name(netdev);
1469 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1470 update_port(ofproto, netdev_name);
1473 struct ofproto_port ofproto_port;
1475 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1477 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1478 ofproto_port_destroy(&ofproto_port);
1483 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1484 * initializes '*port' appropriately; on failure, returns a positive errno
1487 * The caller owns the data in 'ofproto_port' and must free it with
1488 * ofproto_port_destroy() when it is no longer needed. */
1490 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1491 struct ofproto_port *port)
1495 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1497 memset(port, 0, sizeof *port);
1502 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1503 * Returns 0 if successful, otherwise a positive errno. */
1505 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1507 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1508 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1509 struct simap_node *ofp_request_node;
1512 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1513 if (ofp_request_node) {
1514 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1517 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1518 if (!error && ofport) {
1519 /* 'name' is the netdev's name and update_port() is going to close the
1520 * netdev. Just in case update_port() refers to 'name' after it
1521 * destroys 'ofport', make a copy of it around the update_port()
1523 char *devname = xstrdup(name);
1524 update_port(ofproto, devname);
1530 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1531 * performs the 'n_actions' actions in 'actions'. The new flow will not
1534 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1535 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1536 * controllers; otherwise, it will be hidden.
1538 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1540 * This is a helper function for in-band control and fail-open. */
1542 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1543 unsigned int priority,
1544 const struct ofpact *ofpacts, size_t ofpacts_len)
1546 const struct rule *rule;
1548 rule = rule_from_cls_rule(classifier_find_match_exactly(
1549 &ofproto->tables[0].cls, match, priority));
1550 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1551 ofpacts, ofpacts_len)) {
1552 struct ofputil_flow_mod fm;
1554 memset(&fm, 0, sizeof fm);
1556 fm.priority = priority;
1557 fm.buffer_id = UINT32_MAX;
1558 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1559 fm.ofpacts_len = ofpacts_len;
1560 add_flow(ofproto, NULL, &fm, NULL);
1565 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1566 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1567 * operation cannot be initiated now but may be retried later.
1569 * This is a helper function for in-band control and fail-open. */
1571 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1573 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1576 /* Searches for a rule with matching criteria exactly equal to 'target' in
1577 * ofproto's table 0 and, if it finds one, deletes it.
1579 * This is a helper function for in-band control and fail-open. */
1581 ofproto_delete_flow(struct ofproto *ofproto,
1582 const struct match *target, unsigned int priority)
1586 rule = rule_from_cls_rule(classifier_find_match_exactly(
1587 &ofproto->tables[0].cls, target, priority));
1589 /* No such rule -> success. */
1591 } else if (rule->pending) {
1592 /* An operation on the rule is already pending -> failure.
1593 * Caller must retry later if it's important. */
1596 /* Initiate deletion -> success. */
1597 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1598 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1599 oftable_remove_rule(rule);
1600 ofproto->ofproto_class->rule_destruct(rule);
1601 ofopgroup_submit(group);
1607 /* Starts the process of deleting all of the flows from all of ofproto's flow
1608 * tables and then reintroducing the flows required by in-band control and
1609 * fail-open. The process will complete in a later call to ofproto_run(). */
1611 ofproto_flush_flows(struct ofproto *ofproto)
1613 COVERAGE_INC(ofproto_flush);
1614 ofproto->state = S_FLUSH;
1618 reinit_ports(struct ofproto *p)
1620 struct ofproto_port_dump dump;
1621 struct sset devnames;
1622 struct ofport *ofport;
1623 struct ofproto_port ofproto_port;
1624 const char *devname;
1626 COVERAGE_INC(ofproto_reinit_ports);
1628 sset_init(&devnames);
1629 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1630 sset_add(&devnames, netdev_get_name(ofport->netdev));
1632 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1633 sset_add(&devnames, ofproto_port.name);
1636 SSET_FOR_EACH (devname, &devnames) {
1637 update_port(p, devname);
1639 sset_destroy(&devnames);
1643 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1646 uint16_t end_port_no = ofproto->alloc_port_no;
1648 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1649 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1651 if (ofp_port >= ofproto->max_ports
1652 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1653 /* Search for a free OpenFlow port number. We try not to
1654 * immediately reuse them to prevent problems due to old
1657 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1658 ofproto->alloc_port_no = 0;
1660 if (!bitmap_is_set(ofproto->ofp_port_ids,
1661 ofproto->alloc_port_no)) {
1662 ofp_port = ofproto->alloc_port_no;
1665 if (ofproto->alloc_port_no == end_port_no) {
1670 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1675 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1677 if (ofp_port < ofproto->max_ports) {
1678 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1682 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1683 * pointer if the netdev cannot be opened. On success, also fills in
1685 static struct netdev *
1686 ofport_open(struct ofproto *ofproto,
1687 struct ofproto_port *ofproto_port,
1688 struct ofputil_phy_port *pp)
1690 enum netdev_flags flags;
1691 struct netdev *netdev;
1694 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1696 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1697 "cannot be opened (%s)",
1699 ofproto_port->name, ofproto_port->ofp_port,
1700 ofproto_port->name, strerror(error));
1704 if (ofproto_port->ofp_port == OFPP_NONE) {
1705 if (!strcmp(ofproto->name, ofproto_port->name)) {
1706 ofproto_port->ofp_port = OFPP_LOCAL;
1708 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1709 ofproto_port->name);
1712 pp->port_no = ofproto_port->ofp_port;
1713 netdev_get_etheraddr(netdev, pp->hw_addr);
1714 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1715 netdev_get_flags(netdev, &flags);
1716 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1717 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1718 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1719 &pp->supported, &pp->peer);
1720 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1721 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1726 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1727 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1730 ofport_equal(const struct ofputil_phy_port *a,
1731 const struct ofputil_phy_port *b)
1733 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1734 && a->state == b->state
1735 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1736 && a->curr == b->curr
1737 && a->advertised == b->advertised
1738 && a->supported == b->supported
1739 && a->peer == b->peer
1740 && a->curr_speed == b->curr_speed
1741 && a->max_speed == b->max_speed);
1744 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1745 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1746 * one with the same name or port number). */
1748 ofport_install(struct ofproto *p,
1749 struct netdev *netdev, const struct ofputil_phy_port *pp)
1751 const char *netdev_name = netdev_get_name(netdev);
1752 struct ofport *ofport;
1755 /* Create ofport. */
1756 ofport = p->ofproto_class->port_alloc();
1761 ofport->ofproto = p;
1762 ofport->netdev = netdev;
1763 ofport->change_seq = netdev_change_seq(netdev);
1765 ofport->ofp_port = pp->port_no;
1767 /* Add port to 'p'. */
1768 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1769 shash_add(&p->port_by_name, netdev_name, ofport);
1771 update_mtu(p, ofport);
1773 /* Let the ofproto_class initialize its private data. */
1774 error = p->ofproto_class->port_construct(ofport);
1778 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1782 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1783 p->name, netdev_name, strerror(error));
1785 ofport_destroy__(ofport);
1787 netdev_close(netdev);
1791 /* Removes 'ofport' from 'p' and destroys it. */
1793 ofport_remove(struct ofport *ofport)
1795 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1797 ofport_destroy(ofport);
1800 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1803 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1805 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1807 ofport_remove(port);
1811 /* Updates 'port' with new 'pp' description.
1813 * Does not handle a name or port number change. The caller must implement
1814 * such a change as a delete followed by an add. */
1816 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1818 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1819 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1820 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1821 port->pp.state = pp->state;
1822 port->pp.curr = pp->curr;
1823 port->pp.advertised = pp->advertised;
1824 port->pp.supported = pp->supported;
1825 port->pp.peer = pp->peer;
1826 port->pp.curr_speed = pp->curr_speed;
1827 port->pp.max_speed = pp->max_speed;
1829 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1832 /* Update OpenFlow 'state' in 'port' and notify controller. */
1834 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1836 if (port->pp.state != state) {
1837 port->pp.state = state;
1838 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1844 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1846 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1848 if (port->ofproto->ofproto_class->set_realdev) {
1849 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1851 if (port->ofproto->ofproto_class->set_stp_port) {
1852 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1854 if (port->ofproto->ofproto_class->set_cfm) {
1855 port->ofproto->ofproto_class->set_cfm(port, NULL);
1857 if (port->ofproto->ofproto_class->bundle_remove) {
1858 port->ofproto->ofproto_class->bundle_remove(port);
1864 ofport_destroy__(struct ofport *port)
1866 struct ofproto *ofproto = port->ofproto;
1867 const char *name = netdev_get_name(port->netdev);
1869 hmap_remove(&ofproto->ports, &port->hmap_node);
1870 shash_delete(&ofproto->port_by_name,
1871 shash_find(&ofproto->port_by_name, name));
1873 netdev_close(port->netdev);
1874 ofproto->ofproto_class->port_dealloc(port);
1878 ofport_destroy(struct ofport *port)
1881 dealloc_ofp_port(port->ofproto, port->ofp_port);
1882 port->ofproto->ofproto_class->port_destruct(port);
1883 ofport_destroy__(port);
1888 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1890 struct ofport *port;
1892 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1893 hash_int(ofp_port, 0), &ofproto->ports) {
1894 if (port->ofp_port == ofp_port) {
1902 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1904 struct ofproto *ofproto = port->ofproto;
1907 if (ofproto->ofproto_class->port_get_stats) {
1908 error = ofproto->ofproto_class->port_get_stats(port, stats);
1917 update_port(struct ofproto *ofproto, const char *name)
1919 struct ofproto_port ofproto_port;
1920 struct ofputil_phy_port pp;
1921 struct netdev *netdev;
1922 struct ofport *port;
1924 COVERAGE_INC(ofproto_update_port);
1926 /* Fetch 'name''s location and properties from the datapath. */
1927 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1928 ? ofport_open(ofproto, &ofproto_port, &pp)
1931 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1932 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1933 struct netdev *old_netdev = port->netdev;
1935 /* 'name' hasn't changed location. Any properties changed? */
1936 if (!ofport_equal(&port->pp, &pp)) {
1937 ofport_modified(port, &pp);
1940 update_mtu(ofproto, port);
1942 /* Install the newly opened netdev in case it has changed.
1943 * Don't close the old netdev yet in case port_modified has to
1944 * remove a retained reference to it.*/
1945 port->netdev = netdev;
1946 port->change_seq = netdev_change_seq(netdev);
1948 if (port->ofproto->ofproto_class->port_modified) {
1949 port->ofproto->ofproto_class->port_modified(port);
1952 netdev_close(old_netdev);
1954 /* If 'port' is nonnull then its name differs from 'name' and thus
1955 * we should delete it. If we think there's a port named 'name'
1956 * then its port number must be wrong now so delete it too. */
1958 ofport_remove(port);
1960 ofport_remove_with_name(ofproto, name);
1961 ofport_install(ofproto, netdev, &pp);
1964 /* Any port named 'name' is gone now. */
1965 ofport_remove_with_name(ofproto, name);
1967 ofproto_port_destroy(&ofproto_port);
1971 init_ports(struct ofproto *p)
1973 struct ofproto_port_dump dump;
1974 struct ofproto_port ofproto_port;
1975 struct shash_node *node, *next;
1977 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1978 const char *name = ofproto_port.name;
1980 if (shash_find(&p->port_by_name, name)) {
1981 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1984 struct ofputil_phy_port pp;
1985 struct netdev *netdev;
1987 /* Check if an OpenFlow port number had been requested. */
1988 node = shash_find(&init_ofp_ports, name);
1990 const struct iface_hint *iface_hint = node->data;
1991 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
1994 netdev = ofport_open(p, &ofproto_port, &pp);
1996 ofport_install(p, netdev, &pp);
2001 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2002 struct iface_hint *iface_hint = node->data;
2004 if (!strcmp(iface_hint->br_name, p->name)) {
2005 free(iface_hint->br_name);
2006 free(iface_hint->br_type);
2008 shash_delete(&init_ofp_ports, node);
2015 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2016 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2018 find_min_mtu(struct ofproto *p)
2020 struct ofport *ofport;
2023 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2024 struct netdev *netdev = ofport->netdev;
2027 /* Skip any internal ports, since that's what we're trying to
2029 if (!strcmp(netdev_get_type(netdev), "internal")) {
2033 if (netdev_get_mtu(netdev, &dev_mtu)) {
2036 if (!mtu || dev_mtu < mtu) {
2041 return mtu ? mtu: ETH_PAYLOAD_MAX;
2044 /* Update MTU of all datapath devices on 'p' to the minimum of the
2045 * non-datapath ports in event of 'port' added or changed. */
2047 update_mtu(struct ofproto *p, struct ofport *port)
2049 struct ofport *ofport;
2050 struct netdev *netdev = port->netdev;
2051 int dev_mtu, old_min;
2053 if (netdev_get_mtu(netdev, &dev_mtu)) {
2057 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2058 if (dev_mtu > p->min_mtu) {
2059 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2060 dev_mtu = p->min_mtu;
2063 port->mtu = dev_mtu;
2067 /* For non-internal port find new min mtu. */
2068 old_min = p->min_mtu;
2069 port->mtu = dev_mtu;
2070 p->min_mtu = find_min_mtu(p);
2071 if (p->min_mtu == old_min) {
2075 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2076 struct netdev *netdev = ofport->netdev;
2078 if (!strcmp(netdev_get_type(netdev), "internal")) {
2079 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2080 ofport->mtu = p->min_mtu;
2087 ofproto_rule_destroy__(struct rule *rule)
2090 cls_rule_destroy(&rule->cr);
2091 free(rule->ofpacts);
2092 rule->ofproto->ofproto_class->rule_dealloc(rule);
2096 /* This function allows an ofproto implementation to destroy any rules that
2097 * remain when its ->destruct() function is called. The caller must have
2098 * already uninitialized any derived members of 'rule' (step 5 described in the
2099 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2100 * This function implements steps 6 and 7.
2102 * This function should only be called from an ofproto implementation's
2103 * ->destruct() function. It is not suitable elsewhere. */
2105 ofproto_rule_destroy(struct rule *rule)
2107 ovs_assert(!rule->pending);
2108 oftable_remove_rule(rule);
2109 ofproto_rule_destroy__(rule);
2112 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2113 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2115 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2117 return (port == OFPP_ANY
2118 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2121 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2122 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2124 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2126 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2131 case OFOPERATION_ADD:
2132 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2134 case OFOPERATION_DELETE:
2137 case OFOPERATION_MODIFY:
2138 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2144 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2145 * statistics appropriately. 'packet' must have at least sizeof(struct
2146 * ofp10_packet_in) bytes of headroom.
2148 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2149 * with statistics for 'packet' either way.
2151 * Takes ownership of 'packet'. */
2153 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2157 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2159 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2160 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2163 /* Returns true if 'rule' should be hidden from the controller.
2165 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2166 * (e.g. by in-band control) and are intentionally hidden from the
2169 ofproto_rule_is_hidden(const struct rule *rule)
2171 return rule->cr.priority > UINT16_MAX;
2174 static enum oftable_flags
2175 rule_get_flags(const struct rule *rule)
2177 return rule->ofproto->tables[rule->table_id].flags;
2181 rule_is_modifiable(const struct rule *rule)
2183 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2187 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2189 ofconn_send_reply(ofconn, make_echo_reply(oh));
2194 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2196 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2197 struct ofputil_switch_features features;
2198 struct ofport *port;
2204 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2206 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2208 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2209 * if present, are always at the end.) */
2210 n_tables = ofproto->n_tables;
2211 for (i = 0; i < ofproto->n_tables; i++) {
2212 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2218 features.datapath_id = ofproto->datapath_id;
2219 features.n_buffers = pktbuf_capacity();
2220 features.n_tables = n_tables;
2221 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2222 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2224 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2226 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2227 features.auxiliary_id = 0;
2228 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2230 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2231 ofputil_put_switch_features_port(&port->pp, b);
2234 ofconn_send_reply(ofconn, b);
2239 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2241 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2242 struct ofp_switch_config *osc;
2243 enum ofp_config_flags flags;
2247 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2248 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2249 flags = ofproto->frag_handling;
2250 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2251 if (oh->version < OFP13_VERSION
2252 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2253 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2255 osc->flags = htons(flags);
2256 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2257 ofconn_send_reply(ofconn, buf);
2263 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2265 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2266 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2267 uint16_t flags = ntohs(osc->flags);
2269 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2270 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2271 enum ofp_config_flags cur = ofproto->frag_handling;
2272 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2274 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2276 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2277 ofproto->frag_handling = next;
2279 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2281 ofputil_frag_handling_to_string(next));
2285 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2286 ofconn_set_invalid_ttl_to_controller(ofconn,
2287 (oh->version < OFP13_VERSION
2288 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2290 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2295 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2296 * error message code for the caller to propagate upward. Otherwise, returns
2299 * The log message mentions 'msg_type'. */
2301 reject_slave_controller(struct ofconn *ofconn)
2303 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2304 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2305 return OFPERR_OFPBRC_EPERM;
2312 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2314 struct ofproto *p = ofconn_get_ofproto(ofconn);
2315 struct ofputil_packet_out po;
2316 struct ofpbuf *payload;
2317 uint64_t ofpacts_stub[1024 / 8];
2318 struct ofpbuf ofpacts;
2322 COVERAGE_INC(ofproto_packet_out);
2324 error = reject_slave_controller(ofconn);
2329 /* Decode message. */
2330 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2331 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2333 goto exit_free_ofpacts;
2335 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2336 error = OFPERR_OFPBRC_BAD_PORT;
2337 goto exit_free_ofpacts;
2342 if (po.buffer_id != UINT32_MAX) {
2343 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2344 if (error || !payload) {
2345 goto exit_free_ofpacts;
2348 payload = xmalloc(sizeof *payload);
2349 ofpbuf_use_const(payload, po.packet, po.packet_len);
2352 /* Verify actions against packet, then send packet if successful. */
2353 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2354 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2356 error = p->ofproto_class->packet_out(p, payload, &flow,
2357 po.ofpacts, po.ofpacts_len);
2359 ofpbuf_delete(payload);
2362 ofpbuf_uninit(&ofpacts);
2368 update_port_config(struct ofport *port,
2369 enum ofputil_port_config config,
2370 enum ofputil_port_config mask)
2372 enum ofputil_port_config old_config = port->pp.config;
2373 enum ofputil_port_config toggle;
2375 toggle = (config ^ port->pp.config) & mask;
2376 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2377 if (config & OFPUTIL_PC_PORT_DOWN) {
2378 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2380 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2382 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2385 port->pp.config ^= toggle;
2386 if (port->pp.config != old_config) {
2387 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2392 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2394 struct ofproto *p = ofconn_get_ofproto(ofconn);
2395 struct ofputil_port_mod pm;
2396 struct ofport *port;
2399 error = reject_slave_controller(ofconn);
2404 error = ofputil_decode_port_mod(oh, &pm);
2409 port = ofproto_get_port(p, pm.port_no);
2411 return OFPERR_OFPPMFC_BAD_PORT;
2412 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2413 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2415 update_port_config(port, pm.config, pm.mask);
2417 netdev_set_advertisements(port->netdev, pm.advertise);
2424 handle_desc_stats_request(struct ofconn *ofconn,
2425 const struct ofp_header *request)
2427 static const char *default_mfr_desc = "Nicira, Inc.";
2428 static const char *default_hw_desc = "Open vSwitch";
2429 static const char *default_sw_desc = VERSION;
2430 static const char *default_serial_desc = "None";
2431 static const char *default_dp_desc = "None";
2433 struct ofproto *p = ofconn_get_ofproto(ofconn);
2434 struct ofp_desc_stats *ods;
2437 msg = ofpraw_alloc_stats_reply(request, 0);
2438 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2439 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2440 sizeof ods->mfr_desc);
2441 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2442 sizeof ods->hw_desc);
2443 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2444 sizeof ods->sw_desc);
2445 ovs_strlcpy(ods->serial_num,
2446 p->serial_desc ? p->serial_desc : default_serial_desc,
2447 sizeof ods->serial_num);
2448 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2449 sizeof ods->dp_desc);
2450 ofconn_send_reply(ofconn, msg);
2456 handle_table_stats_request(struct ofconn *ofconn,
2457 const struct ofp_header *request)
2459 struct ofproto *p = ofconn_get_ofproto(ofconn);
2460 struct ofp12_table_stats *ots;
2465 /* Set up default values.
2467 * ofp12_table_stats is used as a generic structure as
2468 * it is able to hold all the fields for ofp10_table_stats
2469 * and ofp11_table_stats (and of course itself).
2471 ots = xcalloc(p->n_tables, sizeof *ots);
2472 for (i = 0; i < p->n_tables; i++) {
2473 ots[i].table_id = i;
2474 sprintf(ots[i].name, "table%zu", i);
2475 ots[i].match = htonll(OFPXMT12_MASK);
2476 ots[i].wildcards = htonll(OFPXMT12_MASK);
2477 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2478 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2479 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2480 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2481 ots[i].metadata_match = htonll(UINT64_MAX);
2482 ots[i].metadata_write = htonll(UINT64_MAX);
2483 ots[i].instructions = htonl(OFPIT11_ALL);
2484 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2485 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2486 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2489 p->ofproto_class->get_tables(p, ots);
2491 /* Post-process the tables, dropping hidden tables. */
2492 n_tables = p->n_tables;
2493 for (i = 0; i < p->n_tables; i++) {
2494 const struct oftable *table = &p->tables[i];
2496 if (table->flags & OFTABLE_HIDDEN) {
2502 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2505 if (table->max_flows < ntohl(ots[i].max_entries)) {
2506 ots[i].max_entries = htonl(table->max_flows);
2510 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2511 ofconn_send_reply(ofconn, msg);
2519 append_port_stat(struct ofport *port, struct list *replies)
2521 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2523 /* Intentionally ignore return value, since errors will set
2524 * 'stats' to all-1s, which is correct for OpenFlow, and
2525 * netdev_get_stats() will log errors. */
2526 ofproto_port_get_stats(port, &ops.stats);
2528 ofputil_append_port_stat(replies, &ops);
2532 handle_port_stats_request(struct ofconn *ofconn,
2533 const struct ofp_header *request)
2535 struct ofproto *p = ofconn_get_ofproto(ofconn);
2536 struct ofport *port;
2537 struct list replies;
2541 error = ofputil_decode_port_stats_request(request, &port_no);
2546 ofpmp_init(&replies, request);
2547 if (port_no != OFPP_ANY) {
2548 port = ofproto_get_port(p, port_no);
2550 append_port_stat(port, &replies);
2553 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2554 append_port_stat(port, &replies);
2558 ofconn_send_replies(ofconn, &replies);
2563 handle_port_desc_stats_request(struct ofconn *ofconn,
2564 const struct ofp_header *request)
2566 struct ofproto *p = ofconn_get_ofproto(ofconn);
2567 enum ofp_version version;
2568 struct ofport *port;
2569 struct list replies;
2571 ofpmp_init(&replies, request);
2573 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2574 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2575 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2578 ofconn_send_replies(ofconn, &replies);
2583 calc_flow_duration__(long long int start, long long int now,
2584 uint32_t *sec, uint32_t *nsec)
2586 long long int msecs = now - start;
2587 *sec = msecs / 1000;
2588 *nsec = (msecs % 1000) * (1000 * 1000);
2591 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2592 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2594 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2596 return (table_id == 0xff || table_id < ofproto->n_tables
2598 : OFPERR_OFPBRC_BAD_TABLE_ID);
2602 static struct oftable *
2603 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2605 struct oftable *table;
2607 for (table = &ofproto->tables[table_id];
2608 table < &ofproto->tables[ofproto->n_tables];
2610 if (!(table->flags & OFTABLE_HIDDEN)) {
2618 static struct oftable *
2619 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2621 if (table_id == 0xff) {
2622 return next_visible_table(ofproto, 0);
2623 } else if (table_id < ofproto->n_tables) {
2624 return &ofproto->tables[table_id];
2630 static struct oftable *
2631 next_matching_table(const struct ofproto *ofproto,
2632 const struct oftable *table, uint8_t table_id)
2634 return (table_id == 0xff
2635 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2639 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2641 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2642 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2644 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2645 * only once, for that table. (This can be used to access tables marked
2648 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2649 * entered at all. (Perhaps you should have validated TABLE_ID with
2650 * check_table_id().)
2652 * All parameters are evaluated multiple times.
2654 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2655 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2657 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2659 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2660 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2661 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2664 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2665 * to 'out_port' are included.
2667 * Hidden rules are always omitted.
2669 * Returns 0 on success, otherwise an OpenFlow error code. */
2671 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2672 const struct match *match,
2673 ovs_be64 cookie, ovs_be64 cookie_mask,
2674 uint16_t out_port, struct list *rules)
2676 struct oftable *table;
2680 error = check_table_id(ofproto, table_id);
2686 cls_rule_init(&cr, match, 0);
2687 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2688 struct cls_cursor cursor;
2691 cls_cursor_init(&cursor, &table->cls, &cr);
2692 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2693 if (rule->pending) {
2694 error = OFPROTO_POSTPONE;
2697 if (!ofproto_rule_is_hidden(rule)
2698 && ofproto_rule_has_out_port(rule, out_port)
2699 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2700 list_push_back(rules, &rule->ofproto_node);
2706 cls_rule_destroy(&cr);
2710 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2711 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2712 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2715 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2716 * to 'out_port' are included.
2718 * Hidden rules are always omitted.
2720 * Returns 0 on success, otherwise an OpenFlow error code. */
2722 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2723 const struct match *match, unsigned int priority,
2724 ovs_be64 cookie, ovs_be64 cookie_mask,
2725 uint16_t out_port, struct list *rules)
2727 struct oftable *table;
2731 error = check_table_id(ofproto, table_id);
2737 cls_rule_init(&cr, match, priority);
2738 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2741 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2744 if (rule->pending) {
2745 error = OFPROTO_POSTPONE;
2748 if (!ofproto_rule_is_hidden(rule)
2749 && ofproto_rule_has_out_port(rule, out_port)
2750 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2751 list_push_back(rules, &rule->ofproto_node);
2757 cls_rule_destroy(&cr);
2761 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2762 * forced into the range of a uint16_t. */
2764 age_secs(long long int age_ms)
2766 return (age_ms < 0 ? 0
2767 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2768 : (unsigned int) age_ms / 1000);
2772 handle_flow_stats_request(struct ofconn *ofconn,
2773 const struct ofp_header *request)
2775 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2776 struct ofputil_flow_stats_request fsr;
2777 struct list replies;
2782 error = ofputil_decode_flow_stats_request(&fsr, request);
2787 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2788 fsr.cookie, fsr.cookie_mask,
2789 fsr.out_port, &rules);
2794 ofpmp_init(&replies, request);
2795 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2796 long long int now = time_msec();
2797 struct ofputil_flow_stats fs;
2799 minimatch_expand(&rule->cr.match, &fs.match);
2800 fs.priority = rule->cr.priority;
2801 fs.cookie = rule->flow_cookie;
2802 fs.table_id = rule->table_id;
2803 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2805 fs.idle_timeout = rule->idle_timeout;
2806 fs.hard_timeout = rule->hard_timeout;
2807 fs.idle_age = age_secs(now - rule->used);
2808 fs.hard_age = age_secs(now - rule->modified);
2809 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2811 fs.ofpacts = rule->ofpacts;
2812 fs.ofpacts_len = rule->ofpacts_len;
2814 if (rule->send_flow_removed) {
2815 fs.flags |= OFPFF_SEND_FLOW_REM;
2816 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2817 and OFPFF13_NO_BYT_COUNTS */
2819 ofputil_append_flow_stats_reply(&fs, &replies);
2821 ofconn_send_replies(ofconn, &replies);
2827 flow_stats_ds(struct rule *rule, struct ds *results)
2829 uint64_t packet_count, byte_count;
2831 rule->ofproto->ofproto_class->rule_get_stats(rule,
2832 &packet_count, &byte_count);
2834 if (rule->table_id != 0) {
2835 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2837 ds_put_format(results, "duration=%llds, ",
2838 (time_msec() - rule->created) / 1000);
2839 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2840 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2841 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2842 cls_rule_format(&rule->cr, results);
2843 ds_put_char(results, ',');
2844 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2845 ds_put_cstr(results, "\n");
2848 /* Adds a pretty-printed description of all flows to 'results', including
2849 * hidden flows (e.g., set up by in-band control). */
2851 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2853 struct oftable *table;
2855 OFPROTO_FOR_EACH_TABLE (table, p) {
2856 struct cls_cursor cursor;
2859 cls_cursor_init(&cursor, &table->cls, NULL);
2860 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2861 flow_stats_ds(rule, results);
2866 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2867 * '*engine_type' and '*engine_id', respectively. */
2869 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2870 uint8_t *engine_type, uint8_t *engine_id)
2872 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2875 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2876 * true if the port's CFM status was successfully stored into '*status'.
2877 * Returns false if the port did not have CFM configured, in which case
2878 * '*status' is indeterminate.
2880 * The caller must provide and owns '*status', but it does not own and must not
2881 * modify or free the array returned in 'status->rmps'. */
2883 ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
2884 struct ofproto_cfm_status *status)
2886 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2888 && ofproto->ofproto_class->get_cfm_status
2889 && ofproto->ofproto_class->get_cfm_status(ofport, status));
2893 handle_aggregate_stats_request(struct ofconn *ofconn,
2894 const struct ofp_header *oh)
2896 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2897 struct ofputil_flow_stats_request request;
2898 struct ofputil_aggregate_stats stats;
2899 bool unknown_packets, unknown_bytes;
2900 struct ofpbuf *reply;
2905 error = ofputil_decode_flow_stats_request(&request, oh);
2910 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2911 request.cookie, request.cookie_mask,
2912 request.out_port, &rules);
2917 memset(&stats, 0, sizeof stats);
2918 unknown_packets = unknown_bytes = false;
2919 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2920 uint64_t packet_count;
2921 uint64_t byte_count;
2923 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2926 if (packet_count == UINT64_MAX) {
2927 unknown_packets = true;
2929 stats.packet_count += packet_count;
2932 if (byte_count == UINT64_MAX) {
2933 unknown_bytes = true;
2935 stats.byte_count += byte_count;
2940 if (unknown_packets) {
2941 stats.packet_count = UINT64_MAX;
2943 if (unknown_bytes) {
2944 stats.byte_count = UINT64_MAX;
2947 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2948 ofconn_send_reply(ofconn, reply);
2953 struct queue_stats_cbdata {
2954 struct ofport *ofport;
2955 struct list replies;
2959 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2960 const struct netdev_queue_stats *stats)
2963 struct ofputil_queue_stats oqs = {
2964 .port_no = cbdata->ofport->pp.port_no,
2965 .queue_id = queue_id,
2968 ofputil_append_queue_stat(&cbdata->replies, &oqs);
2972 handle_queue_stats_dump_cb(uint32_t queue_id,
2973 struct netdev_queue_stats *stats,
2976 struct queue_stats_cbdata *cbdata = cbdata_;
2978 put_queue_stats(cbdata, queue_id, stats);
2982 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2983 struct queue_stats_cbdata *cbdata)
2985 cbdata->ofport = port;
2986 if (queue_id == OFPQ_ALL) {
2987 netdev_dump_queue_stats(port->netdev,
2988 handle_queue_stats_dump_cb, cbdata);
2990 struct netdev_queue_stats stats;
2992 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2993 put_queue_stats(cbdata, queue_id, &stats);
2995 return OFPERR_OFPQOFC_BAD_QUEUE;
3002 handle_queue_stats_request(struct ofconn *ofconn,
3003 const struct ofp_header *rq)
3005 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3006 struct queue_stats_cbdata cbdata;
3007 struct ofport *port;
3009 struct ofputil_queue_stats_request oqsr;
3011 COVERAGE_INC(ofproto_queue_req);
3013 ofpmp_init(&cbdata.replies, rq);
3015 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3020 if (oqsr.port_no == OFPP_ANY) {
3021 error = OFPERR_OFPQOFC_BAD_QUEUE;
3022 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3023 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3028 port = ofproto_get_port(ofproto, oqsr.port_no);
3030 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3031 : OFPERR_OFPQOFC_BAD_PORT);
3034 ofconn_send_replies(ofconn, &cbdata.replies);
3036 ofpbuf_list_delete(&cbdata.replies);
3043 is_flow_deletion_pending(const struct ofproto *ofproto,
3044 const struct cls_rule *cls_rule,
3047 if (!hmap_is_empty(&ofproto->deletions)) {
3048 struct ofoperation *op;
3050 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3051 cls_rule_hash(cls_rule, table_id),
3052 &ofproto->deletions) {
3053 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3062 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3063 * in which no matching flow already exists in the flow table.
3065 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3066 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3067 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3068 * initiated now but may be retried later.
3070 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3071 * ownership remains with the caller.
3073 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3076 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3077 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3079 struct oftable *table;
3080 struct ofopgroup *group;
3081 struct rule *victim;
3086 error = check_table_id(ofproto, fm->table_id);
3092 if (fm->table_id == 0xff) {
3094 if (ofproto->ofproto_class->rule_choose_table) {
3095 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3101 ovs_assert(table_id < ofproto->n_tables);
3102 table = &ofproto->tables[table_id];
3104 table = &ofproto->tables[0];
3106 } else if (fm->table_id < ofproto->n_tables) {
3107 table = &ofproto->tables[fm->table_id];
3109 return OFPERR_OFPBRC_BAD_TABLE_ID;
3112 if (table->flags & OFTABLE_READONLY) {
3113 return OFPERR_OFPBRC_EPERM;
3116 /* Allocate new rule and initialize classifier rule. */
3117 rule = ofproto->ofproto_class->rule_alloc();
3119 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3120 ofproto->name, strerror(error));
3123 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3125 /* Serialize against pending deletion. */
3126 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3127 cls_rule_destroy(&rule->cr);
3128 ofproto->ofproto_class->rule_dealloc(rule);
3129 return OFPROTO_POSTPONE;
3132 /* Check for overlap, if requested. */
3133 if (fm->flags & OFPFF_CHECK_OVERLAP
3134 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3135 cls_rule_destroy(&rule->cr);
3136 ofproto->ofproto_class->rule_dealloc(rule);
3137 return OFPERR_OFPFMFC_OVERLAP;
3140 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3142 rule->ofproto = ofproto;
3143 rule->pending = NULL;
3144 rule->flow_cookie = fm->new_cookie;
3145 rule->created = rule->modified = rule->used = time_msec();
3146 rule->idle_timeout = fm->idle_timeout;
3147 rule->hard_timeout = fm->hard_timeout;
3148 rule->table_id = table - ofproto->tables;
3149 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3150 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3151 and OFPFF13_NO_BYT_COUNTS */
3152 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3153 rule->ofpacts_len = fm->ofpacts_len;
3154 rule->evictable = true;
3155 rule->eviction_group = NULL;
3156 list_init(&rule->expirable);
3157 rule->monitor_flags = 0;
3158 rule->add_seqno = 0;
3159 rule->modify_seqno = 0;
3161 /* Insert new rule. */
3162 victim = oftable_replace_rule(rule);
3163 if (victim && !rule_is_modifiable(victim)) {
3164 error = OFPERR_OFPBRC_EPERM;
3165 } else if (victim && victim->pending) {
3166 error = OFPROTO_POSTPONE;
3168 struct ofoperation *op;
3171 if (classifier_count(&table->cls) > table->max_flows) {
3174 was_evictable = rule->evictable;
3175 rule->evictable = false;
3176 evict = choose_rule_to_evict(table);
3177 rule->evictable = was_evictable;
3180 error = OFPERR_OFPFMFC_TABLE_FULL;
3182 } else if (evict->pending) {
3183 error = OFPROTO_POSTPONE;
3190 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3191 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3192 op->victim = victim;
3194 error = ofproto->ofproto_class->rule_construct(rule);
3196 op->group->n_running--;
3197 ofoperation_destroy(rule->pending);
3199 delete_flow__(evict, group);
3201 ofopgroup_submit(group);
3205 /* Back out if an error occurred. */
3207 oftable_substitute_rule(rule, victim);
3208 ofproto_rule_destroy__(rule);
3213 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3215 /* Modifies the rules listed in 'rules', changing their actions to match those
3218 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3221 * Returns 0 on success, otherwise an OpenFlow error code. */
3223 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3224 const struct ofputil_flow_mod *fm,
3225 const struct ofp_header *request, struct list *rules)
3227 struct ofopgroup *group;
3231 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3232 error = OFPERR_OFPBRC_EPERM;
3233 LIST_FOR_EACH (rule, ofproto_node, rules) {
3234 struct ofoperation *op;
3235 bool actions_changed;
3236 ovs_be64 new_cookie;
3238 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3240 if (rule_is_modifiable(rule)) {
3241 /* At least one rule is modifiable, don't report EPERM error. */
3247 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3248 rule->ofpacts, rule->ofpacts_len);
3249 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3251 : rule->flow_cookie);
3253 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3254 rule->flow_cookie = new_cookie;
3255 if (actions_changed) {
3256 op->ofpacts = rule->ofpacts;
3257 op->ofpacts_len = rule->ofpacts_len;
3258 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3259 rule->ofpacts_len = fm->ofpacts_len;
3260 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3262 ofoperation_complete(op, 0);
3265 ofopgroup_submit(group);
3271 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3272 const struct ofputil_flow_mod *fm,
3273 const struct ofp_header *request)
3275 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3278 return add_flow(ofproto, ofconn, fm, request);
3281 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3284 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3287 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3288 const struct ofputil_flow_mod *fm,
3289 const struct ofp_header *request)
3294 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3295 fm->cookie, fm->cookie_mask,
3299 } else if (list_is_empty(&rules)) {
3300 return modify_flows_add(ofproto, ofconn, fm, request);
3302 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3306 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3309 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3312 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3313 const struct ofputil_flow_mod *fm,
3314 const struct ofp_header *request)
3319 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3320 fm->priority, fm->cookie, fm->cookie_mask,
3325 } else if (list_is_empty(&rules)) {
3326 return modify_flows_add(ofproto, ofconn, fm, request);
3328 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3329 fm, request, &rules)
3334 /* OFPFC_DELETE implementation. */
3337 delete_flow__(struct rule *rule, struct ofopgroup *group)
3339 struct ofproto *ofproto = rule->ofproto;
3341 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3343 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3344 oftable_remove_rule(rule);
3345 ofproto->ofproto_class->rule_destruct(rule);
3348 /* Deletes the rules listed in 'rules'.
3350 * Returns 0 on success, otherwise an OpenFlow error code. */
3352 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3353 const struct ofp_header *request, struct list *rules)
3355 struct rule *rule, *next;
3356 struct ofopgroup *group;
3358 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3359 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3360 delete_flow__(rule, group);
3362 ofopgroup_submit(group);
3367 /* Implements OFPFC_DELETE. */
3369 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3370 const struct ofputil_flow_mod *fm,
3371 const struct ofp_header *request)
3376 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3377 fm->cookie, fm->cookie_mask,
3378 fm->out_port, &rules);
3379 return (error ? error
3380 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3385 /* Implements OFPFC_DELETE_STRICT. */
3387 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3388 const struct ofputil_flow_mod *fm,
3389 const struct ofp_header *request)
3394 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3395 fm->priority, fm->cookie, fm->cookie_mask,
3396 fm->out_port, &rules);
3397 return (error ? error
3398 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3404 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3406 struct ofputil_flow_removed fr;
3408 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3412 minimatch_expand(&rule->cr.match, &fr.match);
3413 fr.priority = rule->cr.priority;
3414 fr.cookie = rule->flow_cookie;
3416 fr.table_id = rule->table_id;
3417 calc_flow_duration__(rule->created, time_msec(),
3418 &fr.duration_sec, &fr.duration_nsec);
3419 fr.idle_timeout = rule->idle_timeout;
3420 fr.hard_timeout = rule->hard_timeout;
3421 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3424 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3428 ofproto_rule_update_used(struct rule *rule, long long int used)
3430 if (used > rule->used) {
3431 struct eviction_group *evg = rule->eviction_group;
3435 heap_change(&evg->rules, &rule->evg_node,
3436 rule_eviction_priority(rule));
3441 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3442 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3445 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3448 * ofproto implementation ->run() functions should use this function to expire
3449 * OpenFlow flows. */
3451 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3453 struct ofproto *ofproto = rule->ofproto;
3454 struct ofopgroup *group;
3456 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3458 ofproto_rule_send_removed(rule, reason);
3460 group = ofopgroup_create_unattached(ofproto);
3461 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3462 oftable_remove_rule(rule);
3463 ofproto->ofproto_class->rule_destruct(rule);
3464 ofopgroup_submit(group);
3468 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3470 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3471 struct ofputil_flow_mod fm;
3472 uint64_t ofpacts_stub[1024 / 8];
3473 struct ofpbuf ofpacts;
3477 error = reject_slave_controller(ofconn);
3482 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3483 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3486 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3487 &fm.match.flow, ofproto->max_ports);
3490 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3493 goto exit_free_ofpacts;
3496 /* Record the operation for logging a summary report. */
3497 switch (fm.command) {
3503 case OFPFC_MODIFY_STRICT:
3504 ofproto->n_modify++;
3508 case OFPFC_DELETE_STRICT:
3509 ofproto->n_delete++;
3514 if (ofproto->next_op_report == LLONG_MAX) {
3515 ofproto->first_op = now;
3516 ofproto->next_op_report = MAX(now + 10 * 1000,
3517 ofproto->op_backoff);
3518 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3520 ofproto->last_op = now;
3523 ofpbuf_uninit(&ofpacts);
3529 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3530 const struct ofputil_flow_mod *fm,
3531 const struct ofp_header *oh)
3533 if (ofproto->n_pending >= 50) {
3534 ovs_assert(!list_is_empty(&ofproto->pending));
3535 return OFPROTO_POSTPONE;
3538 switch (fm->command) {
3540 return add_flow(ofproto, ofconn, fm, oh);
3543 return modify_flows_loose(ofproto, ofconn, fm, oh);
3545 case OFPFC_MODIFY_STRICT:
3546 return modify_flow_strict(ofproto, ofconn, fm, oh);
3549 return delete_flows_loose(ofproto, ofconn, fm, oh);
3551 case OFPFC_DELETE_STRICT:
3552 return delete_flow_strict(ofproto, ofconn, fm, oh);
3555 if (fm->command > 0xff) {
3556 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3557 "flow_mod_table_id extension is not enabled",
3560 return OFPERR_OFPFMFC_BAD_COMMAND;
3565 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3567 struct ofputil_role_request request;
3568 struct ofputil_role_request reply;
3572 error = ofputil_decode_role_message(oh, &request);
3577 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3578 if (ofconn_get_role(ofconn) != request.role
3579 && ofconn_has_pending_opgroups(ofconn)) {
3580 return OFPROTO_POSTPONE;
3583 if (request.have_generation_id
3584 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3585 return OFPERR_OFPRRFC_STALE;
3588 ofconn_set_role(ofconn, request.role);
3591 reply.role = ofconn_get_role(ofconn);
3592 reply.have_generation_id = ofconn_get_master_election_id(
3593 ofconn, &reply.generation_id);
3594 buf = ofputil_encode_role_reply(oh, &reply);
3595 ofconn_send_reply(ofconn, buf);
3601 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3602 const struct ofp_header *oh)
3604 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3605 enum ofputil_protocol cur, next;
3607 cur = ofconn_get_protocol(ofconn);
3608 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3609 ofconn_set_protocol(ofconn, next);
3615 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3617 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3618 enum ofputil_protocol cur, next;
3619 enum ofputil_protocol next_base;
3621 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3623 return OFPERR_OFPBRC_EPERM;
3626 cur = ofconn_get_protocol(ofconn);
3627 next = ofputil_protocol_set_base(cur, next_base);
3628 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3629 /* Avoid sending async messages in surprising protocol. */
3630 return OFPROTO_POSTPONE;
3633 ofconn_set_protocol(ofconn, next);
3638 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3639 const struct ofp_header *oh)
3641 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3644 format = ntohl(msg->format);
3645 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3646 return OFPERR_OFPBRC_EPERM;
3649 if (format != ofconn_get_packet_in_format(ofconn)
3650 && ofconn_has_pending_opgroups(ofconn)) {
3651 /* Avoid sending async message in surprsing packet in format. */
3652 return OFPROTO_POSTPONE;
3655 ofconn_set_packet_in_format(ofconn, format);
3660 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3662 const struct nx_async_config *msg = ofpmsg_body(oh);
3663 uint32_t master[OAM_N_TYPES];
3664 uint32_t slave[OAM_N_TYPES];
3666 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3667 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3668 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3670 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3671 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3672 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3674 ofconn_set_async_config(ofconn, master, slave);
3675 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3676 !ofconn_get_miss_send_len(ofconn)) {
3677 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3684 handle_nxt_set_controller_id(struct ofconn *ofconn,
3685 const struct ofp_header *oh)
3687 const struct nx_controller_id *nci = ofpmsg_body(oh);
3689 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3690 return OFPERR_NXBRC_MUST_BE_ZERO;
3693 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3698 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3702 if (ofconn_has_pending_opgroups(ofconn)) {
3703 return OFPROTO_POSTPONE;
3706 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3707 ? OFPRAW_OFPT10_BARRIER_REPLY
3708 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3709 ofconn_send_reply(ofconn, buf);
3714 ofproto_compose_flow_refresh_update(const struct rule *rule,
3715 enum nx_flow_monitor_flags flags,
3718 struct ofoperation *op = rule->pending;
3719 struct ofputil_flow_update fu;
3722 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3723 /* We'll report the final flow when the operation completes. Reporting
3724 * it now would cause a duplicate report later. */
3728 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3729 ? NXFME_ADDED : NXFME_MODIFIED);
3731 fu.idle_timeout = rule->idle_timeout;
3732 fu.hard_timeout = rule->hard_timeout;
3733 fu.table_id = rule->table_id;
3734 fu.cookie = rule->flow_cookie;
3735 minimatch_expand(&rule->cr.match, &match);
3737 fu.priority = rule->cr.priority;
3738 if (!(flags & NXFMF_ACTIONS)) {
3742 fu.ofpacts = rule->ofpacts;
3743 fu.ofpacts_len = rule->ofpacts_len;
3745 /* An operation is in progress. Use the previous version of the flow's
3746 * actions, so that when the operation commits we report the change. */
3748 case OFOPERATION_ADD:
3749 /* We already verified that there was a victim. */
3750 fu.ofpacts = op->victim->ofpacts;
3751 fu.ofpacts_len = op->victim->ofpacts_len;
3754 case OFOPERATION_MODIFY:
3756 fu.ofpacts = op->ofpacts;
3757 fu.ofpacts_len = op->ofpacts_len;
3759 fu.ofpacts = rule->ofpacts;
3760 fu.ofpacts_len = rule->ofpacts_len;
3764 case OFOPERATION_DELETE:
3765 fu.ofpacts = rule->ofpacts;
3766 fu.ofpacts_len = rule->ofpacts_len;
3774 if (list_is_empty(msgs)) {
3775 ofputil_start_flow_update(msgs);
3777 ofputil_append_flow_update(&fu, msgs);
3781 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3785 LIST_FOR_EACH (rule, ofproto_node, rules) {
3786 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3787 rule->monitor_flags = 0;
3789 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3794 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3795 struct rule *rule, uint64_t seqno,
3798 enum nx_flow_monitor_flags update;
3800 if (ofproto_rule_is_hidden(rule)) {
3805 ? ofoperation_has_out_port(rule->pending, m->out_port)
3806 : ofproto_rule_has_out_port(rule, m->out_port))) {
3811 if (rule->add_seqno > seqno) {
3812 update = NXFMF_ADD | NXFMF_MODIFY;
3813 } else if (rule->modify_seqno > seqno) {
3814 update = NXFMF_MODIFY;
3819 if (!(m->flags & update)) {
3823 update = NXFMF_INITIAL;
3826 if (!rule->monitor_flags) {
3827 list_push_back(rules, &rule->ofproto_node);
3829 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3833 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3837 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3838 const struct ofoperation *op;
3839 const struct oftable *table;
3840 struct cls_rule target;
3842 cls_rule_init_from_minimatch(&target, &m->match, 0);
3843 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3844 struct cls_cursor cursor;
3847 cls_cursor_init(&cursor, &table->cls, &target);
3848 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3849 ovs_assert(!rule->pending); /* XXX */
3850 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3854 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3855 struct rule *rule = op->rule;
3857 if (((m->table_id == 0xff
3858 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3859 : m->table_id == rule->table_id))
3860 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3861 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3864 cls_rule_destroy(&target);
3868 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3871 if (m->flags & NXFMF_INITIAL) {
3872 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3877 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3878 uint64_t seqno, struct list *rules)
3880 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3884 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3886 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3887 struct ofmonitor **monitors;
3888 size_t n_monitors, allocated_monitors;
3889 struct list replies;
3896 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3898 n_monitors = allocated_monitors = 0;
3900 struct ofputil_flow_monitor_request request;
3901 struct ofmonitor *m;
3904 retval = ofputil_decode_flow_monitor_request(&request, &b);
3905 if (retval == EOF) {
3907 } else if (retval) {
3912 if (request.table_id != 0xff
3913 && request.table_id >= ofproto->n_tables) {
3914 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3918 error = ofmonitor_create(&request, ofconn, &m);
3923 if (n_monitors >= allocated_monitors) {
3924 monitors = x2nrealloc(monitors, &allocated_monitors,
3927 monitors[n_monitors++] = m;
3931 for (i = 0; i < n_monitors; i++) {
3932 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3935 ofpmp_init(&replies, oh);
3936 ofmonitor_compose_refresh_updates(&rules, &replies);
3937 ofconn_send_replies(ofconn, &replies);
3944 for (i = 0; i < n_monitors; i++) {
3945 ofmonitor_destroy(monitors[i]);
3952 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3954 struct ofmonitor *m;
3957 id = ofputil_decode_flow_monitor_cancel(oh);
3958 m = ofmonitor_lookup(ofconn, id);
3960 return OFPERR_NXBRC_FM_BAD_ID;
3963 ofmonitor_destroy(m);
3968 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3970 const struct ofp_header *oh = msg->data;
3974 error = ofptype_decode(&type, oh);
3980 /* OpenFlow requests. */
3981 case OFPTYPE_ECHO_REQUEST:
3982 return handle_echo_request(ofconn, oh);
3984 case OFPTYPE_FEATURES_REQUEST:
3985 return handle_features_request(ofconn, oh);
3987 case OFPTYPE_GET_CONFIG_REQUEST:
3988 return handle_get_config_request(ofconn, oh);
3990 case OFPTYPE_SET_CONFIG:
3991 return handle_set_config(ofconn, oh);
3993 case OFPTYPE_PACKET_OUT:
3994 return handle_packet_out(ofconn, oh);
3996 case OFPTYPE_PORT_MOD:
3997 return handle_port_mod(ofconn, oh);
3999 case OFPTYPE_FLOW_MOD:
4000 return handle_flow_mod(ofconn, oh);
4002 case OFPTYPE_BARRIER_REQUEST:
4003 return handle_barrier_request(ofconn, oh);
4005 case OFPTYPE_ROLE_REQUEST:
4006 return handle_role_request(ofconn, oh);
4008 /* OpenFlow replies. */
4009 case OFPTYPE_ECHO_REPLY:
4012 /* Nicira extension requests. */
4013 case OFPTYPE_FLOW_MOD_TABLE_ID:
4014 return handle_nxt_flow_mod_table_id(ofconn, oh);
4016 case OFPTYPE_SET_FLOW_FORMAT:
4017 return handle_nxt_set_flow_format(ofconn, oh);
4019 case OFPTYPE_SET_PACKET_IN_FORMAT:
4020 return handle_nxt_set_packet_in_format(ofconn, oh);
4022 case OFPTYPE_SET_CONTROLLER_ID:
4023 return handle_nxt_set_controller_id(ofconn, oh);
4025 case OFPTYPE_FLOW_AGE:
4026 /* Nothing to do. */
4029 case OFPTYPE_FLOW_MONITOR_CANCEL:
4030 return handle_flow_monitor_cancel(ofconn, oh);
4032 case OFPTYPE_SET_ASYNC_CONFIG:
4033 return handle_nxt_set_async_config(ofconn, oh);
4035 /* Statistics requests. */
4036 case OFPTYPE_DESC_STATS_REQUEST:
4037 return handle_desc_stats_request(ofconn, oh);
4039 case OFPTYPE_FLOW_STATS_REQUEST:
4040 return handle_flow_stats_request(ofconn, oh);
4042 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4043 return handle_aggregate_stats_request(ofconn, oh);
4045 case OFPTYPE_TABLE_STATS_REQUEST:
4046 return handle_table_stats_request(ofconn, oh);
4048 case OFPTYPE_PORT_STATS_REQUEST:
4049 return handle_port_stats_request(ofconn, oh);
4051 case OFPTYPE_QUEUE_STATS_REQUEST:
4052 return handle_queue_stats_request(ofconn, oh);
4054 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4055 return handle_port_desc_stats_request(ofconn, oh);
4057 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4058 return handle_flow_monitor_request(ofconn, oh);
4060 /* FIXME: Change the following once they are implemented: */
4061 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4062 case OFPTYPE_GET_ASYNC_REQUEST:
4063 case OFPTYPE_METER_MOD:
4064 case OFPTYPE_GROUP_REQUEST:
4065 case OFPTYPE_GROUP_DESC_REQUEST:
4066 case OFPTYPE_GROUP_FEATURES_REQUEST:
4067 case OFPTYPE_METER_REQUEST:
4068 case OFPTYPE_METER_CONFIG_REQUEST:
4069 case OFPTYPE_METER_FEATURES_REQUEST:
4070 case OFPTYPE_TABLE_FEATURES_REQUEST:
4071 return OFPERR_OFPBRC_BAD_TYPE;
4075 case OFPTYPE_FEATURES_REPLY:
4076 case OFPTYPE_GET_CONFIG_REPLY:
4077 case OFPTYPE_PACKET_IN:
4078 case OFPTYPE_FLOW_REMOVED:
4079 case OFPTYPE_PORT_STATUS:
4080 case OFPTYPE_BARRIER_REPLY:
4081 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4082 case OFPTYPE_DESC_STATS_REPLY:
4083 case OFPTYPE_FLOW_STATS_REPLY:
4084 case OFPTYPE_QUEUE_STATS_REPLY:
4085 case OFPTYPE_PORT_STATS_REPLY:
4086 case OFPTYPE_TABLE_STATS_REPLY:
4087 case OFPTYPE_AGGREGATE_STATS_REPLY:
4088 case OFPTYPE_PORT_DESC_STATS_REPLY:
4089 case OFPTYPE_ROLE_REPLY:
4090 case OFPTYPE_FLOW_MONITOR_PAUSED:
4091 case OFPTYPE_FLOW_MONITOR_RESUMED:
4092 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4093 case OFPTYPE_GET_ASYNC_REPLY:
4094 case OFPTYPE_GROUP_REPLY:
4095 case OFPTYPE_GROUP_DESC_REPLY:
4096 case OFPTYPE_GROUP_FEATURES_REPLY:
4097 case OFPTYPE_METER_REPLY:
4098 case OFPTYPE_METER_CONFIG_REPLY:
4099 case OFPTYPE_METER_FEATURES_REPLY:
4100 case OFPTYPE_TABLE_FEATURES_REPLY:
4102 return OFPERR_OFPBRC_BAD_TYPE;
4107 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4109 int error = handle_openflow__(ofconn, ofp_msg);
4110 if (error && error != OFPROTO_POSTPONE) {
4111 ofconn_send_error(ofconn, ofp_msg->data, error);
4113 COVERAGE_INC(ofproto_recv_openflow);
4114 return error != OFPROTO_POSTPONE;
4117 /* Asynchronous operations. */
4119 /* Creates and returns a new ofopgroup that is not associated with any
4120 * OpenFlow connection.
4122 * The caller should add operations to the returned group with
4123 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4124 static struct ofopgroup *
4125 ofopgroup_create_unattached(struct ofproto *ofproto)
4127 struct ofopgroup *group = xzalloc(sizeof *group);
4128 group->ofproto = ofproto;
4129 list_init(&group->ofproto_node);
4130 list_init(&group->ops);
4131 list_init(&group->ofconn_node);
4135 /* Creates and returns a new ofopgroup for 'ofproto'.
4137 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4138 * connection. The 'request' and 'buffer_id' arguments are ignored.
4140 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4141 * If the ofopgroup eventually fails, then the error reply will include
4142 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4143 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4145 * The caller should add operations to the returned group with
4146 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4147 static struct ofopgroup *
4148 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4149 const struct ofp_header *request, uint32_t buffer_id)
4151 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4153 size_t request_len = ntohs(request->length);
4155 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4157 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4158 group->ofconn = ofconn;
4159 group->request = xmemdup(request, MIN(request_len, 64));
4160 group->buffer_id = buffer_id;
4165 /* Submits 'group' for processing.
4167 * If 'group' contains no operations (e.g. none were ever added, or all of the
4168 * ones that were added completed synchronously), then it is destroyed
4169 * immediately. Otherwise it is added to the ofproto's list of pending
4172 ofopgroup_submit(struct ofopgroup *group)
4174 if (!group->n_running) {
4175 ofopgroup_complete(group);
4177 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4178 group->ofproto->n_pending++;
4183 ofopgroup_complete(struct ofopgroup *group)
4185 struct ofproto *ofproto = group->ofproto;
4187 struct ofconn *abbrev_ofconn;
4188 ovs_be32 abbrev_xid;
4190 struct ofoperation *op, *next_op;
4193 ovs_assert(!group->n_running);
4196 LIST_FOR_EACH (op, group_node, &group->ops) {
4203 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4204 LIST_FOR_EACH (op, group_node, &group->ops) {
4205 if (op->type != OFOPERATION_DELETE) {
4206 struct ofpbuf *packet;
4209 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4213 error = rule_execute(op->rule, in_port, packet);
4220 if (!error && !list_is_empty(&group->ofconn_node)) {
4221 abbrev_ofconn = group->ofconn;
4222 abbrev_xid = group->request->xid;
4224 abbrev_ofconn = NULL;
4225 abbrev_xid = htonl(0);
4227 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4228 struct rule *rule = op->rule;
4230 /* We generally want to report the change to active OpenFlow flow
4231 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4233 - The operation failed.
4235 - The affected rule is not visible to controllers.
4237 - The operation's only effect was to update rule->modified. */
4239 || ofproto_rule_is_hidden(rule)
4240 || (op->type == OFOPERATION_MODIFY
4242 && rule->flow_cookie == op->flow_cookie))) {
4243 /* Check that we can just cast from ofoperation_type to
4244 * nx_flow_update_event. */
4245 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4247 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4249 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4252 ofmonitor_report(ofproto->connmgr, rule,
4253 (enum nx_flow_update_event) op->type,
4254 op->reason, abbrev_ofconn, abbrev_xid);
4257 rule->pending = NULL;
4260 case OFOPERATION_ADD:
4264 ofproto_rule_destroy__(op->victim);
4265 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4266 if (vid_mask == VLAN_VID_MASK) {
4267 if (ofproto->vlan_bitmap) {
4268 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4269 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4270 bitmap_set1(ofproto->vlan_bitmap, vid);
4271 ofproto->vlans_changed = true;
4274 ofproto->vlans_changed = true;
4278 oftable_substitute_rule(rule, op->victim);
4279 ofproto_rule_destroy__(rule);
4283 case OFOPERATION_DELETE:
4284 ovs_assert(!op->error);
4285 ofproto_rule_destroy__(rule);
4289 case OFOPERATION_MODIFY:
4291 rule->modified = time_msec();
4293 rule->flow_cookie = op->flow_cookie;
4295 free(rule->ofpacts);
4296 rule->ofpacts = op->ofpacts;
4297 rule->ofpacts_len = op->ofpacts_len;
4299 op->ofpacts_len = 0;
4308 ofoperation_destroy(op);
4311 ofmonitor_flush(ofproto->connmgr);
4313 if (!list_is_empty(&group->ofproto_node)) {
4314 ovs_assert(ofproto->n_pending > 0);
4315 ofproto->n_pending--;
4316 list_remove(&group->ofproto_node);
4318 if (!list_is_empty(&group->ofconn_node)) {
4319 list_remove(&group->ofconn_node);
4321 ofconn_send_error(group->ofconn, group->request, error);
4323 connmgr_retry(ofproto->connmgr);
4325 free(group->request);
4329 /* Initiates a new operation on 'rule', of the specified 'type', within
4330 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4332 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4333 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4335 * Returns the newly created ofoperation (which is also available as
4336 * rule->pending). */
4337 static struct ofoperation *
4338 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4339 enum ofoperation_type type,
4340 enum ofp_flow_removed_reason reason)
4342 struct ofproto *ofproto = group->ofproto;
4343 struct ofoperation *op;
4345 ovs_assert(!rule->pending);
4347 op = rule->pending = xzalloc(sizeof *op);
4349 list_push_back(&group->ops, &op->group_node);
4352 op->reason = reason;
4353 op->flow_cookie = rule->flow_cookie;
4357 if (type == OFOPERATION_DELETE) {
4358 hmap_insert(&ofproto->deletions, &op->hmap_node,
4359 cls_rule_hash(&rule->cr, rule->table_id));
4366 ofoperation_destroy(struct ofoperation *op)
4368 struct ofopgroup *group = op->group;
4371 op->rule->pending = NULL;
4373 if (op->type == OFOPERATION_DELETE) {
4374 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4376 list_remove(&op->group_node);
4381 /* Indicates that 'op' completed with status 'error', which is either 0 to
4382 * indicate success or an OpenFlow error code on failure.
4384 * If 'error' is 0, indicating success, the operation will be committed
4385 * permanently to the flow table. There is one interesting subcase:
4387 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4388 * the flow table (the "victim" rule) by a new one, then the caller must
4389 * have uninitialized any derived state in the victim rule, as in step 5 in
4390 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4391 * performs steps 6 and 7 for the victim rule, most notably by calling its
4392 * ->rule_dealloc() function.
4394 * If 'error' is nonzero, then generally the operation will be rolled back:
4396 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4397 * restores the original rule. The caller must have uninitialized any
4398 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4399 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4400 * and 7 for the new rule, calling its ->rule_dealloc() function.
4402 * - If 'op' is a "modify flow" operation, ofproto restores the original
4405 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4406 * allowed to fail. It must always succeed.
4408 * Please see the large comment in ofproto/ofproto-provider.h titled
4409 * "Asynchronous Operation Support" for more information. */
4411 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4413 struct ofopgroup *group = op->group;
4415 ovs_assert(op->rule->pending == op);
4416 ovs_assert(group->n_running > 0);
4417 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4420 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4421 ofopgroup_complete(group);
4426 ofoperation_get_victim(struct ofoperation *op)
4428 ovs_assert(op->type == OFOPERATION_ADD);
4433 pick_datapath_id(const struct ofproto *ofproto)
4435 const struct ofport *port;
4437 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4439 uint8_t ea[ETH_ADDR_LEN];
4442 error = netdev_get_etheraddr(port->netdev, ea);
4444 return eth_addr_to_uint64(ea);
4446 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4447 ofproto->name, netdev_get_name(port->netdev),
4450 return ofproto->fallback_dpid;
4454 pick_fallback_dpid(void)
4456 uint8_t ea[ETH_ADDR_LEN];
4457 eth_addr_nicira_random(ea);
4458 return eth_addr_to_uint64(ea);
4461 /* Table overflow policy. */
4463 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4464 * is not configured to evict rules or if the table contains no evictable
4465 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4467 static struct rule *
4468 choose_rule_to_evict(struct oftable *table)
4470 struct eviction_group *evg;
4472 if (!table->eviction_fields) {
4476 /* In the common case, the outer and inner loops here will each be entered
4479 * - The inner loop normally "return"s in its first iteration. If the
4480 * eviction group has any evictable rules, then it always returns in
4483 * - The outer loop only iterates more than once if the largest eviction
4484 * group has no evictable rules.
4486 * - The outer loop can exit only if table's 'max_flows' is all filled up
4487 * by unevictable rules'. */
4488 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4491 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4492 if (rule->evictable) {
4501 /* Searches 'ofproto' for tables that have more flows than their configured
4502 * maximum and that have flow eviction enabled, and evicts as many flows as
4503 * necessary and currently feasible from them.
4505 * This triggers only when an OpenFlow table has N flows in it and then the
4506 * client configures a maximum number of flows less than N. */
4508 ofproto_evict(struct ofproto *ofproto)
4510 struct ofopgroup *group;
4511 struct oftable *table;
4513 group = ofopgroup_create_unattached(ofproto);
4514 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4515 while (classifier_count(&table->cls) > table->max_flows
4516 && table->eviction_fields) {
4519 rule = choose_rule_to_evict(table);
4520 if (!rule || rule->pending) {
4524 ofoperation_create(group, rule,
4525 OFOPERATION_DELETE, OFPRR_EVICTION);
4526 oftable_remove_rule(rule);
4527 ofproto->ofproto_class->rule_destruct(rule);
4530 ofopgroup_submit(group);
4533 /* Eviction groups. */
4535 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4536 * rules. The priority contains low-order random bits to ensure that eviction
4537 * groups with the same number of rules are prioritized randomly. */
4539 eviction_group_priority(size_t n_rules)
4541 uint16_t size = MIN(UINT16_MAX, n_rules);
4542 return (size << 16) | random_uint16();
4545 /* Updates 'evg', an eviction_group within 'table', following a change that
4546 * adds or removes rules in 'evg'. */
4548 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4550 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4551 eviction_group_priority(heap_count(&evg->rules)));
4554 /* Destroys 'evg', an eviction_group within 'table':
4556 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4557 * rules themselves, just removes them from the eviction group.)
4559 * - Removes 'evg' from 'table'.
4563 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4565 while (!heap_is_empty(&evg->rules)) {
4568 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4569 rule->eviction_group = NULL;
4571 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4572 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4573 heap_destroy(&evg->rules);
4577 /* Removes 'rule' from its eviction group, if any. */
4579 eviction_group_remove_rule(struct rule *rule)
4581 if (rule->eviction_group) {
4582 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4583 struct eviction_group *evg = rule->eviction_group;
4585 rule->eviction_group = NULL;
4586 heap_remove(&evg->rules, &rule->evg_node);
4587 if (heap_is_empty(&evg->rules)) {
4588 eviction_group_destroy(table, evg);
4590 eviction_group_resized(table, evg);
4595 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4596 * returns the hash value. */
4598 eviction_group_hash_rule(struct rule *rule)
4600 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4601 const struct mf_subfield *sf;
4605 hash = table->eviction_group_id_basis;
4606 miniflow_expand(&rule->cr.match.flow, &flow);
4607 for (sf = table->eviction_fields;
4608 sf < &table->eviction_fields[table->n_eviction_fields];
4611 if (mf_are_prereqs_ok(sf->field, &flow)) {
4612 union mf_value value;
4614 mf_get_value(sf->field, &flow, &value);
4616 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4618 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4619 unsigned int start = sf->ofs + sf->n_bits;
4620 bitwise_zero(&value, sf->field->n_bytes, start,
4621 sf->field->n_bytes * 8 - start);
4623 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4625 hash = hash_int(hash, 0);
4632 /* Returns an eviction group within 'table' with the given 'id', creating one
4634 static struct eviction_group *
4635 eviction_group_find(struct oftable *table, uint32_t id)
4637 struct eviction_group *evg;
4639 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4643 evg = xmalloc(sizeof *evg);
4644 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4645 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4646 eviction_group_priority(0));
4647 heap_init(&evg->rules);
4652 /* Returns an eviction priority for 'rule'. The return value should be
4653 * interpreted so that higher priorities make a rule more attractive candidates
4656 rule_eviction_priority(struct rule *rule)
4658 long long int hard_expiration;
4659 long long int idle_expiration;
4660 long long int expiration;
4661 uint32_t expiration_offset;
4663 /* Calculate time of expiration. */
4664 hard_expiration = (rule->hard_timeout
4665 ? rule->modified + rule->hard_timeout * 1000
4667 idle_expiration = (rule->idle_timeout
4668 ? rule->used + rule->idle_timeout * 1000
4670 expiration = MIN(hard_expiration, idle_expiration);
4671 if (expiration == LLONG_MAX) {
4675 /* Calculate the time of expiration as a number of (approximate) seconds
4676 * after program startup.
4678 * This should work OK for program runs that last UINT32_MAX seconds or
4679 * less. Therefore, please restart OVS at least once every 136 years. */
4680 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4682 /* Invert the expiration offset because we're using a max-heap. */
4683 return UINT32_MAX - expiration_offset;
4686 /* Adds 'rule' to an appropriate eviction group for its oftable's
4687 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4688 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4691 * The caller must ensure that 'rule' is not already in an eviction group. */
4693 eviction_group_add_rule(struct rule *rule)
4695 struct ofproto *ofproto = rule->ofproto;
4696 struct oftable *table = &ofproto->tables[rule->table_id];
4698 if (table->eviction_fields
4699 && (rule->hard_timeout || rule->idle_timeout)) {
4700 struct eviction_group *evg;
4702 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4704 rule->eviction_group = evg;
4705 heap_insert(&evg->rules, &rule->evg_node,
4706 rule_eviction_priority(rule));
4707 eviction_group_resized(table, evg);
4713 /* Initializes 'table'. */
4715 oftable_init(struct oftable *table)
4717 memset(table, 0, sizeof *table);
4718 classifier_init(&table->cls);
4719 table->max_flows = UINT_MAX;
4722 /* Destroys 'table', including its classifier and eviction groups.
4724 * The caller is responsible for freeing 'table' itself. */
4726 oftable_destroy(struct oftable *table)
4728 ovs_assert(classifier_is_empty(&table->cls));
4729 oftable_disable_eviction(table);
4730 classifier_destroy(&table->cls);
4734 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4735 * string, then 'table' will use its default name.
4737 * This only affects the name exposed for a table exposed through the OpenFlow
4738 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4740 oftable_set_name(struct oftable *table, const char *name)
4742 if (name && name[0]) {
4743 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4744 if (!table->name || strncmp(name, table->name, len)) {
4746 table->name = xmemdup0(name, len);
4754 /* oftables support a choice of two policies when adding a rule would cause the
4755 * number of flows in the table to exceed the configured maximum number: either
4756 * they can refuse to add the new flow or they can evict some existing flow.
4757 * This function configures the former policy on 'table'. */
4759 oftable_disable_eviction(struct oftable *table)
4761 if (table->eviction_fields) {
4762 struct eviction_group *evg, *next;
4764 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4765 &table->eviction_groups_by_id) {
4766 eviction_group_destroy(table, evg);
4768 hmap_destroy(&table->eviction_groups_by_id);
4769 heap_destroy(&table->eviction_groups_by_size);
4771 free(table->eviction_fields);
4772 table->eviction_fields = NULL;
4773 table->n_eviction_fields = 0;
4777 /* oftables support a choice of two policies when adding a rule would cause the
4778 * number of flows in the table to exceed the configured maximum number: either
4779 * they can refuse to add the new flow or they can evict some existing flow.
4780 * This function configures the latter policy on 'table', with fairness based
4781 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4782 * 'n_fields' as 0 disables fairness.) */
4784 oftable_enable_eviction(struct oftable *table,
4785 const struct mf_subfield *fields, size_t n_fields)
4787 struct cls_cursor cursor;
4790 if (table->eviction_fields
4791 && n_fields == table->n_eviction_fields
4793 || !memcmp(fields, table->eviction_fields,
4794 n_fields * sizeof *fields))) {
4799 oftable_disable_eviction(table);
4801 table->n_eviction_fields = n_fields;
4802 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4804 table->eviction_group_id_basis = random_uint32();
4805 hmap_init(&table->eviction_groups_by_id);
4806 heap_init(&table->eviction_groups_by_size);
4808 cls_cursor_init(&cursor, &table->cls, NULL);
4809 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4810 eviction_group_add_rule(rule);
4814 /* Removes 'rule' from the oftable that contains it. */
4816 oftable_remove_rule(struct rule *rule)
4818 struct ofproto *ofproto = rule->ofproto;
4819 struct oftable *table = &ofproto->tables[rule->table_id];
4821 classifier_remove(&table->cls, &rule->cr);
4822 eviction_group_remove_rule(rule);
4823 if (!list_is_empty(&rule->expirable)) {
4824 list_remove(&rule->expirable);
4828 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4829 * oftable that has an identical cls_rule. Returns the rule that was removed,
4830 * if any, and otherwise NULL. */
4831 static struct rule *
4832 oftable_replace_rule(struct rule *rule)
4834 struct ofproto *ofproto = rule->ofproto;
4835 struct oftable *table = &ofproto->tables[rule->table_id];
4836 struct rule *victim;
4837 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4840 list_insert(&ofproto->expirable, &rule->expirable);
4843 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4845 if (!list_is_empty(&victim->expirable)) {
4846 list_remove(&victim->expirable);
4848 eviction_group_remove_rule(victim);
4850 eviction_group_add_rule(rule);
4854 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4856 oftable_substitute_rule(struct rule *old, struct rule *new)
4859 oftable_replace_rule(new);
4861 oftable_remove_rule(old);
4865 /* unixctl commands. */
4868 ofproto_lookup(const char *name)
4870 struct ofproto *ofproto;
4872 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4874 if (!strcmp(ofproto->name, name)) {
4882 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4883 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4885 struct ofproto *ofproto;
4889 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4890 ds_put_format(&results, "%s\n", ofproto->name);
4892 unixctl_command_reply(conn, ds_cstr(&results));
4893 ds_destroy(&results);
4897 ofproto_unixctl_init(void)
4899 static bool registered;
4905 unixctl_command_register("ofproto/list", "", 0, 0,
4906 ofproto_unixctl_list, NULL);
4909 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4911 * This is deprecated. It is only for compatibility with broken device drivers
4912 * in old versions of Linux that do not properly support VLANs when VLAN
4913 * devices are not used. When broken device drivers are no longer in
4914 * widespread use, we will delete these interfaces. */
4916 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4917 * (exactly) by an OpenFlow rule in 'ofproto'. */
4919 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4921 const struct oftable *oftable;
4923 free(ofproto->vlan_bitmap);
4924 ofproto->vlan_bitmap = bitmap_allocate(4096);
4925 ofproto->vlans_changed = false;
4927 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4928 const struct cls_table *table;
4930 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4931 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4932 const struct cls_rule *rule;
4934 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4935 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4936 bitmap_set1(vlan_bitmap, vid);
4937 bitmap_set1(ofproto->vlan_bitmap, vid);
4944 /* Returns true if new VLANs have come into use by the flow table since the
4945 * last call to ofproto_get_vlan_usage().
4947 * We don't track when old VLANs stop being used. */
4949 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4951 return ofproto->vlans_changed;
4954 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4955 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4956 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4957 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4958 * then the VLAN device is un-enslaved. */
4960 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4961 uint16_t realdev_ofp_port, int vid)
4963 struct ofport *ofport;
4966 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
4968 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4970 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4971 ofproto->name, vlandev_ofp_port);
4975 if (!ofproto->ofproto_class->set_realdev) {
4976 if (!vlandev_ofp_port) {
4979 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4983 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4985 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4986 ofproto->name, vlandev_ofp_port,
4987 netdev_get_name(ofport->netdev), strerror(error));