2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
208 static uint64_t pick_datapath_id(const struct ofproto *);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto *);
211 static void update_mtu(struct ofproto *, struct ofport *);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class **ofproto_classes;
218 static size_t n_ofproto_classes;
219 static size_t allocated_ofproto_classes;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
224 /* Initial mappings of port to OpenFlow number mappings. */
225 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
227 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
229 /* Must be called to initialize the ofproto library.
231 * The caller may pass in 'iface_hints', which contains an shash of
232 * "iface_hint" elements indexed by the interface's name. The provider
233 * may use these hints to describe the startup configuration in order to
234 * reinitialize its state. The caller owns the provided data, so a
235 * provider will make copies of anything required. An ofproto provider
236 * will remove any existing state that is not described by the hint, and
237 * may choose to remove it all. */
239 ofproto_init(const struct shash *iface_hints)
241 struct shash_node *node;
244 ofproto_class_register(&ofproto_dpif_class);
246 /* Make a local copy, since we don't own 'iface_hints' elements. */
247 SHASH_FOR_EACH(node, iface_hints) {
248 const struct iface_hint *orig_hint = node->data;
249 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
250 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
252 new_hint->br_name = xstrdup(orig_hint->br_name);
253 new_hint->br_type = xstrdup(br_type);
254 new_hint->ofp_port = orig_hint->ofp_port;
256 shash_add(&init_ofp_ports, node->name, new_hint);
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->init(&init_ofp_ports);
264 /* 'type' should be a normalized datapath type, as returned by
265 * ofproto_normalize_type(). Returns the corresponding ofproto_class
266 * structure, or a null pointer if there is none registered for 'type'. */
267 static const struct ofproto_class *
268 ofproto_class_find__(const char *type)
272 for (i = 0; i < n_ofproto_classes; i++) {
273 const struct ofproto_class *class = ofproto_classes[i];
278 class->enumerate_types(&types);
279 found = sset_contains(&types, type);
280 sset_destroy(&types);
286 VLOG_WARN("unknown datapath type %s", type);
290 /* Registers a new ofproto class. After successful registration, new ofprotos
291 * of that type can be created using ofproto_create(). */
293 ofproto_class_register(const struct ofproto_class *new_class)
297 for (i = 0; i < n_ofproto_classes; i++) {
298 if (ofproto_classes[i] == new_class) {
303 if (n_ofproto_classes >= allocated_ofproto_classes) {
304 ofproto_classes = x2nrealloc(ofproto_classes,
305 &allocated_ofproto_classes,
306 sizeof *ofproto_classes);
308 ofproto_classes[n_ofproto_classes++] = new_class;
312 /* Unregisters a datapath provider. 'type' must have been previously
313 * registered and not currently be in use by any ofprotos. After
314 * unregistration new datapaths of that type cannot be opened using
315 * ofproto_create(). */
317 ofproto_class_unregister(const struct ofproto_class *class)
321 for (i = 0; i < n_ofproto_classes; i++) {
322 if (ofproto_classes[i] == class) {
323 for (i++; i < n_ofproto_classes; i++) {
324 ofproto_classes[i - 1] = ofproto_classes[i];
330 VLOG_WARN("attempted to unregister an ofproto class that is not "
335 /* Clears 'types' and enumerates all registered ofproto types into it. The
336 * caller must first initialize the sset. */
338 ofproto_enumerate_types(struct sset *types)
342 for (i = 0; i < n_ofproto_classes; i++) {
343 ofproto_classes[i]->enumerate_types(types);
347 /* Returns the fully spelled out name for the given ofproto 'type'.
349 * Normalized type string can be compared with strcmp(). Unnormalized type
350 * string might be the same even if they have different spellings. */
352 ofproto_normalize_type(const char *type)
354 return type && type[0] ? type : "system";
357 /* Clears 'names' and enumerates the names of all known created ofprotos with
358 * the given 'type'. The caller must first initialize the sset. Returns 0 if
359 * successful, otherwise a positive errno value.
361 * Some kinds of datapaths might not be practically enumerable. This is not
362 * considered an error. */
364 ofproto_enumerate_names(const char *type, struct sset *names)
366 const struct ofproto_class *class = ofproto_class_find__(type);
367 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
371 ofproto_create(const char *datapath_name, const char *datapath_type,
372 struct ofproto **ofprotop)
374 const struct ofproto_class *class;
375 struct ofproto *ofproto;
381 ofproto_unixctl_init();
383 datapath_type = ofproto_normalize_type(datapath_type);
384 class = ofproto_class_find__(datapath_type);
386 VLOG_WARN("could not create datapath %s of unknown type %s",
387 datapath_name, datapath_type);
391 ofproto = class->alloc();
393 VLOG_ERR("failed to allocate datapath %s of type %s",
394 datapath_name, datapath_type);
399 memset(ofproto, 0, sizeof *ofproto);
400 ofproto->ofproto_class = class;
401 ofproto->name = xstrdup(datapath_name);
402 ofproto->type = xstrdup(datapath_type);
403 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
404 hash_string(ofproto->name, 0));
405 ofproto->datapath_id = 0;
406 ofproto_set_flow_eviction_threshold(ofproto,
407 OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT);
408 ofproto->forward_bpdu = false;
409 ofproto->fallback_dpid = pick_fallback_dpid();
410 ofproto->mfr_desc = NULL;
411 ofproto->hw_desc = NULL;
412 ofproto->sw_desc = NULL;
413 ofproto->serial_desc = NULL;
414 ofproto->dp_desc = NULL;
415 ofproto->frag_handling = OFPC_FRAG_NORMAL;
416 hmap_init(&ofproto->ports);
417 shash_init(&ofproto->port_by_name);
418 simap_init(&ofproto->ofp_requests);
419 ofproto->max_ports = OFPP_MAX;
420 ofproto->tables = NULL;
421 ofproto->n_tables = 0;
422 list_init(&ofproto->expirable);
423 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
424 ofproto->state = S_OPENFLOW;
425 list_init(&ofproto->pending);
426 ofproto->n_pending = 0;
427 hmap_init(&ofproto->deletions);
428 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
429 ofproto->first_op = ofproto->last_op = LLONG_MIN;
430 ofproto->next_op_report = LLONG_MAX;
431 ofproto->op_backoff = LLONG_MIN;
432 ofproto->vlan_bitmap = NULL;
433 ofproto->vlans_changed = false;
434 ofproto->min_mtu = INT_MAX;
436 error = ofproto->ofproto_class->construct(ofproto);
438 VLOG_ERR("failed to open datapath %s: %s",
439 datapath_name, strerror(error));
440 ofproto_destroy__(ofproto);
444 /* The "max_ports" member should have been set by ->construct(ofproto).
445 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
446 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
447 bitmap_set1(ofproto->ofp_port_ids, 0);
449 /* Check that hidden tables, if any, are at the end. */
450 ovs_assert(ofproto->n_tables);
451 for (i = 0; i + 1 < ofproto->n_tables; i++) {
452 enum oftable_flags flags = ofproto->tables[i].flags;
453 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
455 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
458 ofproto->datapath_id = pick_datapath_id(ofproto);
465 /* Must be called (only) by an ofproto implementation in its constructor
466 * function. See the large comment on 'construct' in struct ofproto_class for
469 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
471 struct oftable *table;
473 ovs_assert(!ofproto->n_tables);
474 ovs_assert(n_tables >= 1 && n_tables <= 255);
476 ofproto->n_tables = n_tables;
477 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
478 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
483 /* To be optionally called (only) by an ofproto implementation in its
484 * constructor function. See the large comment on 'construct' in struct
485 * ofproto_class for details.
487 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
488 * will then ensure that actions passed into the ofproto implementation will
489 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
490 * function is not called, there will be no such restriction.
492 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
493 * the 'max_ports' restriction. */
495 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
497 ovs_assert(max_ports <= OFPP_MAX);
498 ofproto->max_ports = max_ports;
502 ofproto_get_datapath_id(const struct ofproto *ofproto)
504 return ofproto->datapath_id;
508 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
510 uint64_t old_dpid = p->datapath_id;
511 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
512 if (p->datapath_id != old_dpid) {
513 /* Force all active connections to reconnect, since there is no way to
514 * notify a controller that the datapath ID has changed. */
515 ofproto_reconnect_controllers(p);
520 ofproto_set_controllers(struct ofproto *p,
521 const struct ofproto_controller *controllers,
522 size_t n_controllers, uint32_t allowed_versions)
524 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
529 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
531 connmgr_set_fail_mode(p->connmgr, fail_mode);
534 /* Drops the connections between 'ofproto' and all of its controllers, forcing
535 * them to reconnect. */
537 ofproto_reconnect_controllers(struct ofproto *ofproto)
539 connmgr_reconnect(ofproto->connmgr);
542 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
543 * in-band control should guarantee access, in the same way that in-band
544 * control guarantees access to OpenFlow controllers. */
546 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
547 const struct sockaddr_in *extras, size_t n)
549 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
552 /* Sets the OpenFlow queue used by flows set up by in-band control on
553 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
554 * flows will use the default queue. */
556 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
558 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
561 /* Sets the number of flows at which eviction from the kernel flow table
564 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
566 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
567 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
569 ofproto->flow_eviction_threshold = threshold;
573 /* If forward_bpdu is true, the NORMAL action will forward frames with
574 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
575 * the NORMAL action will drop these frames. */
577 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
579 bool old_val = ofproto->forward_bpdu;
580 ofproto->forward_bpdu = forward_bpdu;
581 if (old_val != ofproto->forward_bpdu) {
582 if (ofproto->ofproto_class->forward_bpdu_changed) {
583 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
588 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
589 * 'idle_time', in seconds, and the maximum number of MAC table entries to
592 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
595 if (ofproto->ofproto_class->set_mac_table_config) {
596 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
602 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
605 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
609 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
611 return connmgr_set_snoops(ofproto->connmgr, snoops);
615 ofproto_set_netflow(struct ofproto *ofproto,
616 const struct netflow_options *nf_options)
618 if (nf_options && sset_is_empty(&nf_options->collectors)) {
622 if (ofproto->ofproto_class->set_netflow) {
623 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
625 return nf_options ? EOPNOTSUPP : 0;
630 ofproto_set_sflow(struct ofproto *ofproto,
631 const struct ofproto_sflow_options *oso)
633 if (oso && sset_is_empty(&oso->targets)) {
637 if (ofproto->ofproto_class->set_sflow) {
638 return ofproto->ofproto_class->set_sflow(ofproto, oso);
640 return oso ? EOPNOTSUPP : 0;
644 /* Spanning Tree Protocol (STP) configuration. */
646 /* Configures STP on 'ofproto' using the settings defined in 's'. If
647 * 's' is NULL, disables STP.
649 * Returns 0 if successful, otherwise a positive errno value. */
651 ofproto_set_stp(struct ofproto *ofproto,
652 const struct ofproto_stp_settings *s)
654 return (ofproto->ofproto_class->set_stp
655 ? ofproto->ofproto_class->set_stp(ofproto, s)
659 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
660 * 'enabled' member of 's' is false, then the other members are not
663 * Returns 0 if successful, otherwise a positive errno value. */
665 ofproto_get_stp_status(struct ofproto *ofproto,
666 struct ofproto_stp_status *s)
668 return (ofproto->ofproto_class->get_stp_status
669 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
673 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
674 * in 's'. The caller is responsible for assigning STP port numbers
675 * (using the 'port_num' member in the range of 1 through 255, inclusive)
676 * and ensuring there are no duplicates. If the 's' is NULL, then STP
677 * is disabled on the port.
679 * Returns 0 if successful, otherwise a positive errno value.*/
681 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
682 const struct ofproto_port_stp_settings *s)
684 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
686 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
687 ofproto->name, ofp_port);
691 return (ofproto->ofproto_class->set_stp_port
692 ? ofproto->ofproto_class->set_stp_port(ofport, s)
696 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
697 * 's'. If the 'enabled' member in 's' is false, then the other members
698 * are not meaningful.
700 * Returns 0 if successful, otherwise a positive errno value.*/
702 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
703 struct ofproto_port_stp_status *s)
705 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
707 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
708 "port %"PRIu16, ofproto->name, ofp_port);
712 return (ofproto->ofproto_class->get_stp_port_status
713 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
717 /* Queue DSCP configuration. */
719 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
720 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
721 * to implement QoS. Instead, it is used to implement features which require
722 * knowledge of what queues exist on a port, and some basic information about
725 * Returns 0 if successful, otherwise a positive errno value. */
727 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
728 const struct ofproto_port_queue *queues,
731 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
734 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
735 ofproto->name, ofp_port);
739 return (ofproto->ofproto_class->set_queues
740 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
744 /* Connectivity Fault Management configuration. */
746 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
748 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
750 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
751 if (ofport && ofproto->ofproto_class->set_cfm) {
752 ofproto->ofproto_class->set_cfm(ofport, NULL);
756 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
757 * basic configuration from the configuration members in 'cfm', and the remote
758 * maintenance point ID from remote_mpid. Ignores the statistics members of
761 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
763 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
764 const struct cfm_settings *s)
766 struct ofport *ofport;
769 ofport = ofproto_get_port(ofproto, ofp_port);
771 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
772 ofproto->name, ofp_port);
776 /* XXX: For configuration simplicity, we only support one remote_mpid
777 * outside of the CFM module. It's not clear if this is the correct long
778 * term solution or not. */
779 error = (ofproto->ofproto_class->set_cfm
780 ? ofproto->ofproto_class->set_cfm(ofport, s)
783 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
784 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
789 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
790 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
791 * 0 if LACP partner information is not current (generally indicating a
792 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
794 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
796 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
797 return (ofport && ofproto->ofproto_class->port_is_lacp_current
798 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
804 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
805 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
806 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
807 * configuration plus, if there is more than one slave, a bonding
810 * If 'aux' is already registered then this function updates its configuration
811 * to 's'. Otherwise, this function registers a new bundle.
813 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
816 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
817 const struct ofproto_bundle_settings *s)
819 return (ofproto->ofproto_class->bundle_set
820 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
824 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
825 * If no such bundle has been registered, this has no effect. */
827 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
829 return ofproto_bundle_register(ofproto, aux, NULL);
833 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
834 * If 'aux' is already registered then this function updates its configuration
835 * to 's'. Otherwise, this function registers a new mirror. */
837 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
838 const struct ofproto_mirror_settings *s)
840 return (ofproto->ofproto_class->mirror_set
841 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
845 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
846 * If no mirror has been registered, this has no effect. */
848 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
850 return ofproto_mirror_register(ofproto, aux, NULL);
853 /* Retrieves statistics from mirror associated with client data pointer
854 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
855 * 'bytes', respectively. If a particular counters is not supported,
856 * the appropriate argument is set to UINT64_MAX. */
858 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
859 uint64_t *packets, uint64_t *bytes)
861 if (!ofproto->ofproto_class->mirror_get_stats) {
862 *packets = *bytes = UINT64_MAX;
866 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
870 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
871 * which all packets are flooded, instead of using MAC learning. If
872 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
874 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
877 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
879 return (ofproto->ofproto_class->set_flood_vlans
880 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
884 /* Returns true if 'aux' is a registered bundle that is currently in use as the
885 * output for a mirror. */
887 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
889 return (ofproto->ofproto_class->is_mirror_output_bundle
890 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
894 /* Configuration of OpenFlow tables. */
896 /* Returns the number of OpenFlow tables in 'ofproto'. */
898 ofproto_get_n_tables(const struct ofproto *ofproto)
900 return ofproto->n_tables;
903 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
904 * settings from 's'. 'table_id' must be in the range 0 through the number of
905 * OpenFlow tables in 'ofproto' minus 1, inclusive.
907 * For read-only tables, only the name may be configured. */
909 ofproto_configure_table(struct ofproto *ofproto, int table_id,
910 const struct ofproto_table_settings *s)
912 struct oftable *table;
914 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
915 table = &ofproto->tables[table_id];
917 oftable_set_name(table, s->name);
919 if (table->flags & OFTABLE_READONLY) {
924 oftable_enable_eviction(table, s->groups, s->n_groups);
926 oftable_disable_eviction(table);
929 table->max_flows = s->max_flows;
930 if (classifier_count(&table->cls) > table->max_flows
931 && table->eviction_fields) {
932 /* 'table' contains more flows than allowed. We might not be able to
933 * evict them right away because of the asynchronous nature of flow
934 * table changes. Schedule eviction for later. */
935 switch (ofproto->state) {
937 ofproto->state = S_EVICT;
941 /* We're already deleting flows, nothing more to do. */
948 ofproto_has_snoops(const struct ofproto *ofproto)
950 return connmgr_has_snoops(ofproto->connmgr);
954 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
956 connmgr_get_snoops(ofproto->connmgr, snoops);
960 ofproto_flush__(struct ofproto *ofproto)
962 struct ofopgroup *group;
963 struct oftable *table;
965 if (ofproto->ofproto_class->flush) {
966 ofproto->ofproto_class->flush(ofproto);
969 group = ofopgroup_create_unattached(ofproto);
970 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
971 struct rule *rule, *next_rule;
972 struct cls_cursor cursor;
974 if (table->flags & OFTABLE_HIDDEN) {
978 cls_cursor_init(&cursor, &table->cls, NULL);
979 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
980 if (!rule->pending) {
981 ofoperation_create(group, rule, OFOPERATION_DELETE,
983 oftable_remove_rule(rule);
984 ofproto->ofproto_class->rule_destruct(rule);
988 ofopgroup_submit(group);
992 ofproto_destroy__(struct ofproto *ofproto)
994 struct oftable *table;
996 ovs_assert(list_is_empty(&ofproto->pending));
997 ovs_assert(!ofproto->n_pending);
999 connmgr_destroy(ofproto->connmgr);
1001 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1002 free(ofproto->name);
1003 free(ofproto->type);
1004 free(ofproto->mfr_desc);
1005 free(ofproto->hw_desc);
1006 free(ofproto->sw_desc);
1007 free(ofproto->serial_desc);
1008 free(ofproto->dp_desc);
1009 hmap_destroy(&ofproto->ports);
1010 shash_destroy(&ofproto->port_by_name);
1011 bitmap_free(ofproto->ofp_port_ids);
1012 simap_destroy(&ofproto->ofp_requests);
1014 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1015 oftable_destroy(table);
1017 free(ofproto->tables);
1019 hmap_destroy(&ofproto->deletions);
1021 free(ofproto->vlan_bitmap);
1023 ofproto->ofproto_class->dealloc(ofproto);
1027 ofproto_destroy(struct ofproto *p)
1029 struct ofport *ofport, *next_ofport;
1036 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1037 ofport_destroy(ofport);
1040 p->ofproto_class->destruct(p);
1041 ofproto_destroy__(p);
1044 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1045 * kernel datapath, for example, this destroys the datapath in the kernel, and
1046 * with the netdev-based datapath, it tears down the data structures that
1047 * represent the datapath.
1049 * The datapath should not be currently open as an ofproto. */
1051 ofproto_delete(const char *name, const char *type)
1053 const struct ofproto_class *class = ofproto_class_find__(type);
1054 return (!class ? EAFNOSUPPORT
1055 : !class->del ? EACCES
1056 : class->del(type, name));
1060 process_port_change(struct ofproto *ofproto, int error, char *devname)
1062 if (error == ENOBUFS) {
1063 reinit_ports(ofproto);
1064 } else if (!error) {
1065 update_port(ofproto, devname);
1071 ofproto_type_run(const char *datapath_type)
1073 const struct ofproto_class *class;
1076 datapath_type = ofproto_normalize_type(datapath_type);
1077 class = ofproto_class_find__(datapath_type);
1079 error = class->type_run ? class->type_run(datapath_type) : 0;
1080 if (error && error != EAGAIN) {
1081 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1082 datapath_type, strerror(error));
1088 ofproto_type_run_fast(const char *datapath_type)
1090 const struct ofproto_class *class;
1093 datapath_type = ofproto_normalize_type(datapath_type);
1094 class = ofproto_class_find__(datapath_type);
1096 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1097 if (error && error != EAGAIN) {
1098 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1099 datapath_type, strerror(error));
1105 ofproto_type_wait(const char *datapath_type)
1107 const struct ofproto_class *class;
1109 datapath_type = ofproto_normalize_type(datapath_type);
1110 class = ofproto_class_find__(datapath_type);
1112 if (class->type_wait) {
1113 class->type_wait(datapath_type);
1118 ofproto_run(struct ofproto *p)
1120 struct sset changed_netdevs;
1121 const char *changed_netdev;
1122 struct ofport *ofport;
1125 error = p->ofproto_class->run(p);
1126 if (error && error != EAGAIN) {
1127 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1130 if (p->ofproto_class->port_poll) {
1133 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1134 process_port_change(p, error, devname);
1138 /* Update OpenFlow port status for any port whose netdev has changed.
1140 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1141 * destroyed, so it's not safe to update ports directly from the
1142 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1143 * need this two-phase approach. */
1144 sset_init(&changed_netdevs);
1145 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1146 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1147 if (ofport->change_seq != change_seq) {
1148 ofport->change_seq = change_seq;
1149 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1152 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1153 update_port(p, changed_netdev);
1155 sset_destroy(&changed_netdevs);
1159 connmgr_run(p->connmgr, handle_openflow);
1163 connmgr_run(p->connmgr, NULL);
1165 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1166 p->state = S_OPENFLOW;
1171 connmgr_run(p->connmgr, NULL);
1173 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1174 connmgr_flushed(p->connmgr);
1175 p->state = S_OPENFLOW;
1183 if (time_msec() >= p->next_op_report) {
1184 long long int ago = (time_msec() - p->first_op) / 1000;
1185 long long int interval = (p->last_op - p->first_op) / 1000;
1189 ds_put_format(&s, "%d flow_mods ",
1190 p->n_add + p->n_delete + p->n_modify);
1191 if (interval == ago) {
1192 ds_put_format(&s, "in the last %lld s", ago);
1193 } else if (interval) {
1194 ds_put_format(&s, "in the %lld s starting %lld s ago",
1197 ds_put_format(&s, "%lld s ago", ago);
1200 ds_put_cstr(&s, " (");
1202 ds_put_format(&s, "%d adds, ", p->n_add);
1205 ds_put_format(&s, "%d deletes, ", p->n_delete);
1208 ds_put_format(&s, "%d modifications, ", p->n_modify);
1211 ds_put_char(&s, ')');
1213 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1216 p->n_add = p->n_delete = p->n_modify = 0;
1217 p->next_op_report = LLONG_MAX;
1223 /* Performs periodic activity required by 'ofproto' that needs to be done
1224 * with the least possible latency.
1226 * It makes sense to call this function a couple of times per poll loop, to
1227 * provide a significant performance boost on some benchmarks with the
1228 * ofproto-dpif implementation. */
1230 ofproto_run_fast(struct ofproto *p)
1234 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1235 if (error && error != EAGAIN) {
1236 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1237 p->name, strerror(error));
1243 ofproto_wait(struct ofproto *p)
1245 struct ofport *ofport;
1247 p->ofproto_class->wait(p);
1248 if (p->ofproto_class->port_poll_wait) {
1249 p->ofproto_class->port_poll_wait(p);
1252 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1253 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1254 poll_immediate_wake();
1260 connmgr_wait(p->connmgr, true);
1265 connmgr_wait(p->connmgr, false);
1266 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1267 poll_immediate_wake();
1274 ofproto_is_alive(const struct ofproto *p)
1276 return connmgr_has_controllers(p->connmgr);
1279 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1280 * memory_report(). */
1282 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1284 const struct oftable *table;
1285 unsigned int n_rules;
1287 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1288 simap_increase(usage, "ops",
1289 ofproto->n_pending + hmap_count(&ofproto->deletions));
1292 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1293 n_rules += classifier_count(&table->cls);
1295 simap_increase(usage, "rules", n_rules);
1297 if (ofproto->ofproto_class->get_memory_usage) {
1298 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1301 connmgr_get_memory_usage(ofproto->connmgr, usage);
1305 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1308 connmgr_get_controller_info(ofproto->connmgr, info);
1312 ofproto_free_ofproto_controller_info(struct shash *info)
1314 connmgr_free_controller_info(info);
1317 /* Makes a deep copy of 'old' into 'port'. */
1319 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1321 port->name = xstrdup(old->name);
1322 port->type = xstrdup(old->type);
1323 port->ofp_port = old->ofp_port;
1326 /* Frees memory allocated to members of 'ofproto_port'.
1328 * Do not call this function on an ofproto_port obtained from
1329 * ofproto_port_dump_next(): that function retains ownership of the data in the
1332 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1334 free(ofproto_port->name);
1335 free(ofproto_port->type);
1338 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1340 * This function provides no status indication. An error status for the entire
1341 * dump operation is provided when it is completed by calling
1342 * ofproto_port_dump_done().
1345 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1346 const struct ofproto *ofproto)
1348 dump->ofproto = ofproto;
1349 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1353 /* Attempts to retrieve another port from 'dump', which must have been created
1354 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1355 * 'port' and returns true. On failure, returns false.
1357 * Failure might indicate an actual error or merely that the last port has been
1358 * dumped. An error status for the entire dump operation is provided when it
1359 * is completed by calling ofproto_port_dump_done().
1361 * The ofproto owns the data stored in 'port'. It will remain valid until at
1362 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1363 * ofproto_port_dump_done(). */
1365 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1366 struct ofproto_port *port)
1368 const struct ofproto *ofproto = dump->ofproto;
1374 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1377 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1383 /* Completes port table dump operation 'dump', which must have been created
1384 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1385 * error-free, otherwise a positive errno value describing the problem. */
1387 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1389 const struct ofproto *ofproto = dump->ofproto;
1391 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1394 return dump->error == EOF ? 0 : dump->error;
1397 /* Returns the type to pass to netdev_open() when a datapath of type
1398 * 'datapath_type' has a port of type 'port_type', for a few special
1399 * cases when a netdev type differs from a port type. For example, when
1400 * using the userspace datapath, a port of type "internal" needs to be
1403 * Returns either 'type' itself or a string literal, which must not be
1406 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1408 const struct ofproto_class *class;
1410 datapath_type = ofproto_normalize_type(datapath_type);
1411 class = ofproto_class_find__(datapath_type);
1416 return (class->port_open_type
1417 ? class->port_open_type(datapath_type, port_type)
1421 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1422 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1423 * the port's OpenFlow port number.
1425 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1426 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1427 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1428 * 'ofp_portp' is non-null). */
1430 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1431 uint16_t *ofp_portp)
1433 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1436 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1438 const char *netdev_name = netdev_get_name(netdev);
1440 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1441 update_port(ofproto, netdev_name);
1444 struct ofproto_port ofproto_port;
1446 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1448 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1449 ofproto_port_destroy(&ofproto_port);
1454 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1455 * initializes '*port' appropriately; on failure, returns a positive errno
1458 * The caller owns the data in 'ofproto_port' and must free it with
1459 * ofproto_port_destroy() when it is no longer needed. */
1461 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1462 struct ofproto_port *port)
1466 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1468 memset(port, 0, sizeof *port);
1473 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1474 * Returns 0 if successful, otherwise a positive errno. */
1476 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1478 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1479 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1480 struct simap_node *ofp_request_node;
1483 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1484 if (ofp_request_node) {
1485 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1488 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1489 if (!error && ofport) {
1490 /* 'name' is the netdev's name and update_port() is going to close the
1491 * netdev. Just in case update_port() refers to 'name' after it
1492 * destroys 'ofport', make a copy of it around the update_port()
1494 char *devname = xstrdup(name);
1495 update_port(ofproto, devname);
1501 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1502 * performs the 'n_actions' actions in 'actions'. The new flow will not
1505 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1506 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1507 * controllers; otherwise, it will be hidden.
1509 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1511 * This is a helper function for in-band control and fail-open. */
1513 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1514 unsigned int priority,
1515 const struct ofpact *ofpacts, size_t ofpacts_len)
1517 const struct rule *rule;
1519 rule = rule_from_cls_rule(classifier_find_match_exactly(
1520 &ofproto->tables[0].cls, match, priority));
1521 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1522 ofpacts, ofpacts_len)) {
1523 struct ofputil_flow_mod fm;
1525 memset(&fm, 0, sizeof fm);
1527 fm.priority = priority;
1528 fm.buffer_id = UINT32_MAX;
1529 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1530 fm.ofpacts_len = ofpacts_len;
1531 add_flow(ofproto, NULL, &fm, NULL);
1536 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1537 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1538 * operation cannot be initiated now but may be retried later.
1540 * This is a helper function for in-band control and fail-open. */
1542 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1544 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1547 /* Searches for a rule with matching criteria exactly equal to 'target' in
1548 * ofproto's table 0 and, if it finds one, deletes it.
1550 * This is a helper function for in-band control and fail-open. */
1552 ofproto_delete_flow(struct ofproto *ofproto,
1553 const struct match *target, unsigned int priority)
1557 rule = rule_from_cls_rule(classifier_find_match_exactly(
1558 &ofproto->tables[0].cls, target, priority));
1560 /* No such rule -> success. */
1562 } else if (rule->pending) {
1563 /* An operation on the rule is already pending -> failure.
1564 * Caller must retry later if it's important. */
1567 /* Initiate deletion -> success. */
1568 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1569 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1570 oftable_remove_rule(rule);
1571 ofproto->ofproto_class->rule_destruct(rule);
1572 ofopgroup_submit(group);
1578 /* Starts the process of deleting all of the flows from all of ofproto's flow
1579 * tables and then reintroducing the flows required by in-band control and
1580 * fail-open. The process will complete in a later call to ofproto_run(). */
1582 ofproto_flush_flows(struct ofproto *ofproto)
1584 COVERAGE_INC(ofproto_flush);
1585 ofproto->state = S_FLUSH;
1589 reinit_ports(struct ofproto *p)
1591 struct ofproto_port_dump dump;
1592 struct sset devnames;
1593 struct ofport *ofport;
1594 struct ofproto_port ofproto_port;
1595 const char *devname;
1597 COVERAGE_INC(ofproto_reinit_ports);
1599 sset_init(&devnames);
1600 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1601 sset_add(&devnames, netdev_get_name(ofport->netdev));
1603 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1604 sset_add(&devnames, ofproto_port.name);
1607 SSET_FOR_EACH (devname, &devnames) {
1608 update_port(p, devname);
1610 sset_destroy(&devnames);
1614 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1617 uint16_t end_port_no = ofproto->alloc_port_no;
1619 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1620 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1622 if (ofp_port >= ofproto->max_ports
1623 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1624 /* Search for a free OpenFlow port number. We try not to
1625 * immediately reuse them to prevent problems due to old
1628 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1629 ofproto->alloc_port_no = 0;
1631 if (!bitmap_is_set(ofproto->ofp_port_ids,
1632 ofproto->alloc_port_no)) {
1633 ofp_port = ofproto->alloc_port_no;
1636 if (ofproto->alloc_port_no == end_port_no) {
1641 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1646 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1648 if (ofp_port < ofproto->max_ports) {
1649 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1653 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1654 * pointer if the netdev cannot be opened. On success, also fills in
1656 static struct netdev *
1657 ofport_open(struct ofproto *ofproto,
1658 struct ofproto_port *ofproto_port,
1659 struct ofputil_phy_port *pp)
1661 enum netdev_flags flags;
1662 struct netdev *netdev;
1665 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1667 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1668 "cannot be opened (%s)",
1670 ofproto_port->name, ofproto_port->ofp_port,
1671 ofproto_port->name, strerror(error));
1675 if (ofproto_port->ofp_port == OFPP_NONE) {
1676 if (!strcmp(ofproto->name, ofproto_port->name)) {
1677 ofproto_port->ofp_port = OFPP_LOCAL;
1679 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1680 ofproto_port->name);
1683 pp->port_no = ofproto_port->ofp_port;
1684 netdev_get_etheraddr(netdev, pp->hw_addr);
1685 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1686 netdev_get_flags(netdev, &flags);
1687 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1688 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1689 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1690 &pp->supported, &pp->peer);
1691 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1692 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1697 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1698 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1701 ofport_equal(const struct ofputil_phy_port *a,
1702 const struct ofputil_phy_port *b)
1704 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1705 && a->state == b->state
1706 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1707 && a->curr == b->curr
1708 && a->advertised == b->advertised
1709 && a->supported == b->supported
1710 && a->peer == b->peer
1711 && a->curr_speed == b->curr_speed
1712 && a->max_speed == b->max_speed);
1715 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1716 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1717 * one with the same name or port number). */
1719 ofport_install(struct ofproto *p,
1720 struct netdev *netdev, const struct ofputil_phy_port *pp)
1722 const char *netdev_name = netdev_get_name(netdev);
1723 struct ofport *ofport;
1726 /* Create ofport. */
1727 ofport = p->ofproto_class->port_alloc();
1732 ofport->ofproto = p;
1733 ofport->netdev = netdev;
1734 ofport->change_seq = netdev_change_seq(netdev);
1736 ofport->ofp_port = pp->port_no;
1738 /* Add port to 'p'. */
1739 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1740 shash_add(&p->port_by_name, netdev_name, ofport);
1742 update_mtu(p, ofport);
1744 /* Let the ofproto_class initialize its private data. */
1745 error = p->ofproto_class->port_construct(ofport);
1749 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1753 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1754 p->name, netdev_name, strerror(error));
1756 ofport_destroy__(ofport);
1758 netdev_close(netdev);
1762 /* Removes 'ofport' from 'p' and destroys it. */
1764 ofport_remove(struct ofport *ofport)
1766 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1768 ofport_destroy(ofport);
1771 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1774 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1776 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1778 ofport_remove(port);
1782 /* Updates 'port' with new 'pp' description.
1784 * Does not handle a name or port number change. The caller must implement
1785 * such a change as a delete followed by an add. */
1787 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1789 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1790 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1791 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1792 port->pp.state = pp->state;
1793 port->pp.curr = pp->curr;
1794 port->pp.advertised = pp->advertised;
1795 port->pp.supported = pp->supported;
1796 port->pp.peer = pp->peer;
1797 port->pp.curr_speed = pp->curr_speed;
1798 port->pp.max_speed = pp->max_speed;
1800 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1803 /* Update OpenFlow 'state' in 'port' and notify controller. */
1805 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1807 if (port->pp.state != state) {
1808 port->pp.state = state;
1809 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1815 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1817 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1819 if (port->ofproto->ofproto_class->set_realdev) {
1820 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1822 if (port->ofproto->ofproto_class->set_stp_port) {
1823 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1825 if (port->ofproto->ofproto_class->set_cfm) {
1826 port->ofproto->ofproto_class->set_cfm(port, NULL);
1828 if (port->ofproto->ofproto_class->bundle_remove) {
1829 port->ofproto->ofproto_class->bundle_remove(port);
1835 ofport_destroy__(struct ofport *port)
1837 struct ofproto *ofproto = port->ofproto;
1838 const char *name = netdev_get_name(port->netdev);
1840 hmap_remove(&ofproto->ports, &port->hmap_node);
1841 shash_delete(&ofproto->port_by_name,
1842 shash_find(&ofproto->port_by_name, name));
1844 netdev_close(port->netdev);
1845 ofproto->ofproto_class->port_dealloc(port);
1849 ofport_destroy(struct ofport *port)
1852 dealloc_ofp_port(port->ofproto, port->ofp_port);
1853 port->ofproto->ofproto_class->port_destruct(port);
1854 ofport_destroy__(port);
1859 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1861 struct ofport *port;
1863 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1864 hash_int(ofp_port, 0), &ofproto->ports) {
1865 if (port->ofp_port == ofp_port) {
1873 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1875 struct ofproto *ofproto = port->ofproto;
1878 if (ofproto->ofproto_class->port_get_stats) {
1879 error = ofproto->ofproto_class->port_get_stats(port, stats);
1888 update_port(struct ofproto *ofproto, const char *name)
1890 struct ofproto_port ofproto_port;
1891 struct ofputil_phy_port pp;
1892 struct netdev *netdev;
1893 struct ofport *port;
1895 COVERAGE_INC(ofproto_update_port);
1897 /* Fetch 'name''s location and properties from the datapath. */
1898 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1899 ? ofport_open(ofproto, &ofproto_port, &pp)
1902 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1903 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1904 struct netdev *old_netdev = port->netdev;
1906 /* 'name' hasn't changed location. Any properties changed? */
1907 if (!ofport_equal(&port->pp, &pp)) {
1908 ofport_modified(port, &pp);
1911 update_mtu(ofproto, port);
1913 /* Install the newly opened netdev in case it has changed.
1914 * Don't close the old netdev yet in case port_modified has to
1915 * remove a retained reference to it.*/
1916 port->netdev = netdev;
1917 port->change_seq = netdev_change_seq(netdev);
1919 if (port->ofproto->ofproto_class->port_modified) {
1920 port->ofproto->ofproto_class->port_modified(port);
1923 netdev_close(old_netdev);
1925 /* If 'port' is nonnull then its name differs from 'name' and thus
1926 * we should delete it. If we think there's a port named 'name'
1927 * then its port number must be wrong now so delete it too. */
1929 ofport_remove(port);
1931 ofport_remove_with_name(ofproto, name);
1932 ofport_install(ofproto, netdev, &pp);
1935 /* Any port named 'name' is gone now. */
1936 ofport_remove_with_name(ofproto, name);
1938 ofproto_port_destroy(&ofproto_port);
1942 init_ports(struct ofproto *p)
1944 struct ofproto_port_dump dump;
1945 struct ofproto_port ofproto_port;
1946 struct shash_node *node, *next;
1948 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1949 const char *name = ofproto_port.name;
1951 if (shash_find(&p->port_by_name, name)) {
1952 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1955 struct ofputil_phy_port pp;
1956 struct netdev *netdev;
1958 /* Check if an OpenFlow port number had been requested. */
1959 node = shash_find(&init_ofp_ports, name);
1961 const struct iface_hint *iface_hint = node->data;
1962 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
1965 netdev = ofport_open(p, &ofproto_port, &pp);
1967 ofport_install(p, netdev, &pp);
1972 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
1973 struct iface_hint *iface_hint = node->data;
1975 if (!strcmp(iface_hint->br_name, p->name)) {
1976 free(iface_hint->br_name);
1977 free(iface_hint->br_type);
1979 shash_delete(&init_ofp_ports, node);
1986 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1987 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1989 find_min_mtu(struct ofproto *p)
1991 struct ofport *ofport;
1994 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1995 struct netdev *netdev = ofport->netdev;
1998 /* Skip any internal ports, since that's what we're trying to
2000 if (!strcmp(netdev_get_type(netdev), "internal")) {
2004 if (netdev_get_mtu(netdev, &dev_mtu)) {
2007 if (!mtu || dev_mtu < mtu) {
2012 return mtu ? mtu: ETH_PAYLOAD_MAX;
2015 /* Update MTU of all datapath devices on 'p' to the minimum of the
2016 * non-datapath ports in event of 'port' added or changed. */
2018 update_mtu(struct ofproto *p, struct ofport *port)
2020 struct ofport *ofport;
2021 struct netdev *netdev = port->netdev;
2022 int dev_mtu, old_min;
2024 if (netdev_get_mtu(netdev, &dev_mtu)) {
2028 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2029 if (dev_mtu > p->min_mtu) {
2030 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2031 dev_mtu = p->min_mtu;
2034 port->mtu = dev_mtu;
2038 /* For non-internal port find new min mtu. */
2039 old_min = p->min_mtu;
2040 port->mtu = dev_mtu;
2041 p->min_mtu = find_min_mtu(p);
2042 if (p->min_mtu == old_min) {
2046 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2047 struct netdev *netdev = ofport->netdev;
2049 if (!strcmp(netdev_get_type(netdev), "internal")) {
2050 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2051 ofport->mtu = p->min_mtu;
2058 ofproto_rule_destroy__(struct rule *rule)
2061 cls_rule_destroy(&rule->cr);
2062 free(rule->ofpacts);
2063 rule->ofproto->ofproto_class->rule_dealloc(rule);
2067 /* This function allows an ofproto implementation to destroy any rules that
2068 * remain when its ->destruct() function is called. The caller must have
2069 * already uninitialized any derived members of 'rule' (step 5 described in the
2070 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2071 * This function implements steps 6 and 7.
2073 * This function should only be called from an ofproto implementation's
2074 * ->destruct() function. It is not suitable elsewhere. */
2076 ofproto_rule_destroy(struct rule *rule)
2078 ovs_assert(!rule->pending);
2079 oftable_remove_rule(rule);
2080 ofproto_rule_destroy__(rule);
2083 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2084 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2086 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2088 return (port == OFPP_ANY
2089 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2092 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2093 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2095 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2097 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2102 case OFOPERATION_ADD:
2103 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2105 case OFOPERATION_DELETE:
2108 case OFOPERATION_MODIFY:
2109 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2115 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2116 * statistics appropriately. 'packet' must have at least sizeof(struct
2117 * ofp10_packet_in) bytes of headroom.
2119 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2120 * with statistics for 'packet' either way.
2122 * Takes ownership of 'packet'. */
2124 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2128 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2130 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2131 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2134 /* Returns true if 'rule' should be hidden from the controller.
2136 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2137 * (e.g. by in-band control) and are intentionally hidden from the
2140 ofproto_rule_is_hidden(const struct rule *rule)
2142 return rule->cr.priority > UINT16_MAX;
2145 static enum oftable_flags
2146 rule_get_flags(const struct rule *rule)
2148 return rule->ofproto->tables[rule->table_id].flags;
2152 rule_is_modifiable(const struct rule *rule)
2154 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2158 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2160 ofconn_send_reply(ofconn, make_echo_reply(oh));
2165 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2167 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2168 struct ofputil_switch_features features;
2169 struct ofport *port;
2175 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2177 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2179 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2180 * if present, are always at the end.) */
2181 n_tables = ofproto->n_tables;
2182 for (i = 0; i < ofproto->n_tables; i++) {
2183 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2189 features.datapath_id = ofproto->datapath_id;
2190 features.n_buffers = pktbuf_capacity();
2191 features.n_tables = n_tables;
2192 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2193 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2195 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2197 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2198 features.auxiliary_id = 0;
2199 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2201 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2202 ofputil_put_switch_features_port(&port->pp, b);
2205 ofconn_send_reply(ofconn, b);
2210 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2212 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2213 struct ofp_switch_config *osc;
2214 enum ofp_config_flags flags;
2218 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2219 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2220 flags = ofproto->frag_handling;
2221 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2222 if (oh->version < OFP13_VERSION
2223 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2224 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2226 osc->flags = htons(flags);
2227 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2228 ofconn_send_reply(ofconn, buf);
2234 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2236 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2237 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2238 uint16_t flags = ntohs(osc->flags);
2240 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2241 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2242 enum ofp_config_flags cur = ofproto->frag_handling;
2243 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2245 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2247 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2248 ofproto->frag_handling = next;
2250 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2252 ofputil_frag_handling_to_string(next));
2256 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2257 ofconn_set_invalid_ttl_to_controller(ofconn,
2258 (oh->version < OFP13_VERSION
2259 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2261 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2266 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2267 * error message code for the caller to propagate upward. Otherwise, returns
2270 * The log message mentions 'msg_type'. */
2272 reject_slave_controller(struct ofconn *ofconn)
2274 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2275 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2276 return OFPERR_OFPBRC_EPERM;
2283 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2285 struct ofproto *p = ofconn_get_ofproto(ofconn);
2286 struct ofputil_packet_out po;
2287 struct ofpbuf *payload;
2288 uint64_t ofpacts_stub[1024 / 8];
2289 struct ofpbuf ofpacts;
2293 COVERAGE_INC(ofproto_packet_out);
2295 error = reject_slave_controller(ofconn);
2300 /* Decode message. */
2301 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2302 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2304 goto exit_free_ofpacts;
2306 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2307 error = OFPERR_OFPBRC_BAD_PORT;
2308 goto exit_free_ofpacts;
2313 if (po.buffer_id != UINT32_MAX) {
2314 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2315 if (error || !payload) {
2316 goto exit_free_ofpacts;
2319 payload = xmalloc(sizeof *payload);
2320 ofpbuf_use_const(payload, po.packet, po.packet_len);
2323 /* Verify actions against packet, then send packet if successful. */
2324 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2325 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2327 error = p->ofproto_class->packet_out(p, payload, &flow,
2328 po.ofpacts, po.ofpacts_len);
2330 ofpbuf_delete(payload);
2333 ofpbuf_uninit(&ofpacts);
2339 update_port_config(struct ofport *port,
2340 enum ofputil_port_config config,
2341 enum ofputil_port_config mask)
2343 enum ofputil_port_config old_config = port->pp.config;
2344 enum ofputil_port_config toggle;
2346 toggle = (config ^ port->pp.config) & mask;
2347 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2348 if (config & OFPUTIL_PC_PORT_DOWN) {
2349 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2351 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2353 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2356 port->pp.config ^= toggle;
2357 if (port->pp.config != old_config) {
2358 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2363 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2365 struct ofproto *p = ofconn_get_ofproto(ofconn);
2366 struct ofputil_port_mod pm;
2367 struct ofport *port;
2370 error = reject_slave_controller(ofconn);
2375 error = ofputil_decode_port_mod(oh, &pm);
2380 port = ofproto_get_port(p, pm.port_no);
2382 return OFPERR_OFPPMFC_BAD_PORT;
2383 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2384 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2386 update_port_config(port, pm.config, pm.mask);
2388 netdev_set_advertisements(port->netdev, pm.advertise);
2395 handle_desc_stats_request(struct ofconn *ofconn,
2396 const struct ofp_header *request)
2398 static const char *default_mfr_desc = "Nicira, Inc.";
2399 static const char *default_hw_desc = "Open vSwitch";
2400 static const char *default_sw_desc = VERSION;
2401 static const char *default_serial_desc = "None";
2402 static const char *default_dp_desc = "None";
2404 struct ofproto *p = ofconn_get_ofproto(ofconn);
2405 struct ofp_desc_stats *ods;
2408 msg = ofpraw_alloc_stats_reply(request, 0);
2409 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2410 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2411 sizeof ods->mfr_desc);
2412 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2413 sizeof ods->hw_desc);
2414 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2415 sizeof ods->sw_desc);
2416 ovs_strlcpy(ods->serial_num,
2417 p->serial_desc ? p->serial_desc : default_serial_desc,
2418 sizeof ods->serial_num);
2419 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2420 sizeof ods->dp_desc);
2421 ofconn_send_reply(ofconn, msg);
2427 handle_table_stats_request(struct ofconn *ofconn,
2428 const struct ofp_header *request)
2430 struct ofproto *p = ofconn_get_ofproto(ofconn);
2431 struct ofp12_table_stats *ots;
2436 /* Set up default values.
2438 * ofp12_table_stats is used as a generic structure as
2439 * it is able to hold all the fields for ofp10_table_stats
2440 * and ofp11_table_stats (and of course itself).
2442 ots = xcalloc(p->n_tables, sizeof *ots);
2443 for (i = 0; i < p->n_tables; i++) {
2444 ots[i].table_id = i;
2445 sprintf(ots[i].name, "table%zu", i);
2446 ots[i].match = htonll(OFPXMT12_MASK);
2447 ots[i].wildcards = htonll(OFPXMT12_MASK);
2448 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2449 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2450 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2451 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2452 ots[i].metadata_match = htonll(UINT64_MAX);
2453 ots[i].metadata_write = htonll(UINT64_MAX);
2454 ots[i].instructions = htonl(OFPIT11_ALL);
2455 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2456 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2457 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2460 p->ofproto_class->get_tables(p, ots);
2462 /* Post-process the tables, dropping hidden tables. */
2463 n_tables = p->n_tables;
2464 for (i = 0; i < p->n_tables; i++) {
2465 const struct oftable *table = &p->tables[i];
2467 if (table->flags & OFTABLE_HIDDEN) {
2473 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2476 if (table->max_flows < ntohl(ots[i].max_entries)) {
2477 ots[i].max_entries = htonl(table->max_flows);
2481 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2482 ofconn_send_reply(ofconn, msg);
2490 append_port_stat(struct ofport *port, struct list *replies)
2492 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2494 /* Intentionally ignore return value, since errors will set
2495 * 'stats' to all-1s, which is correct for OpenFlow, and
2496 * netdev_get_stats() will log errors. */
2497 ofproto_port_get_stats(port, &ops.stats);
2499 ofputil_append_port_stat(replies, &ops);
2503 handle_port_stats_request(struct ofconn *ofconn,
2504 const struct ofp_header *request)
2506 struct ofproto *p = ofconn_get_ofproto(ofconn);
2507 struct ofport *port;
2508 struct list replies;
2512 error = ofputil_decode_port_stats_request(request, &port_no);
2517 ofpmp_init(&replies, request);
2518 if (port_no != OFPP_ANY) {
2519 port = ofproto_get_port(p, port_no);
2521 append_port_stat(port, &replies);
2524 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2525 append_port_stat(port, &replies);
2529 ofconn_send_replies(ofconn, &replies);
2534 handle_port_desc_stats_request(struct ofconn *ofconn,
2535 const struct ofp_header *request)
2537 struct ofproto *p = ofconn_get_ofproto(ofconn);
2538 enum ofp_version version;
2539 struct ofport *port;
2540 struct list replies;
2542 ofpmp_init(&replies, request);
2544 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2545 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2546 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2549 ofconn_send_replies(ofconn, &replies);
2554 calc_flow_duration__(long long int start, long long int now,
2555 uint32_t *sec, uint32_t *nsec)
2557 long long int msecs = now - start;
2558 *sec = msecs / 1000;
2559 *nsec = (msecs % 1000) * (1000 * 1000);
2562 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2563 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2565 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2567 return (table_id == 0xff || table_id < ofproto->n_tables
2569 : OFPERR_OFPBRC_BAD_TABLE_ID);
2573 static struct oftable *
2574 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2576 struct oftable *table;
2578 for (table = &ofproto->tables[table_id];
2579 table < &ofproto->tables[ofproto->n_tables];
2581 if (!(table->flags & OFTABLE_HIDDEN)) {
2589 static struct oftable *
2590 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2592 if (table_id == 0xff) {
2593 return next_visible_table(ofproto, 0);
2594 } else if (table_id < ofproto->n_tables) {
2595 return &ofproto->tables[table_id];
2601 static struct oftable *
2602 next_matching_table(const struct ofproto *ofproto,
2603 const struct oftable *table, uint8_t table_id)
2605 return (table_id == 0xff
2606 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2610 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2612 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2613 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2615 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2616 * only once, for that table. (This can be used to access tables marked
2619 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2620 * entered at all. (Perhaps you should have validated TABLE_ID with
2621 * check_table_id().)
2623 * All parameters are evaluated multiple times.
2625 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2626 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2628 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2630 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2631 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2632 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2635 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2636 * to 'out_port' are included.
2638 * Hidden rules are always omitted.
2640 * Returns 0 on success, otherwise an OpenFlow error code. */
2642 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2643 const struct match *match,
2644 ovs_be64 cookie, ovs_be64 cookie_mask,
2645 uint16_t out_port, struct list *rules)
2647 struct oftable *table;
2651 error = check_table_id(ofproto, table_id);
2657 cls_rule_init(&cr, match, 0);
2658 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2659 struct cls_cursor cursor;
2662 cls_cursor_init(&cursor, &table->cls, &cr);
2663 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2664 if (rule->pending) {
2665 error = OFPROTO_POSTPONE;
2668 if (!ofproto_rule_is_hidden(rule)
2669 && ofproto_rule_has_out_port(rule, out_port)
2670 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2671 list_push_back(rules, &rule->ofproto_node);
2677 cls_rule_destroy(&cr);
2681 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2682 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2683 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2686 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2687 * to 'out_port' are included.
2689 * Hidden rules are always omitted.
2691 * Returns 0 on success, otherwise an OpenFlow error code. */
2693 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2694 const struct match *match, unsigned int priority,
2695 ovs_be64 cookie, ovs_be64 cookie_mask,
2696 uint16_t out_port, struct list *rules)
2698 struct oftable *table;
2702 error = check_table_id(ofproto, table_id);
2708 cls_rule_init(&cr, match, priority);
2709 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2712 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2715 if (rule->pending) {
2716 error = OFPROTO_POSTPONE;
2719 if (!ofproto_rule_is_hidden(rule)
2720 && ofproto_rule_has_out_port(rule, out_port)
2721 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2722 list_push_back(rules, &rule->ofproto_node);
2728 cls_rule_destroy(&cr);
2732 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2733 * forced into the range of a uint16_t. */
2735 age_secs(long long int age_ms)
2737 return (age_ms < 0 ? 0
2738 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2739 : (unsigned int) age_ms / 1000);
2743 handle_flow_stats_request(struct ofconn *ofconn,
2744 const struct ofp_header *request)
2746 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2747 struct ofputil_flow_stats_request fsr;
2748 struct list replies;
2753 error = ofputil_decode_flow_stats_request(&fsr, request);
2758 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2759 fsr.cookie, fsr.cookie_mask,
2760 fsr.out_port, &rules);
2765 ofpmp_init(&replies, request);
2766 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2767 long long int now = time_msec();
2768 struct ofputil_flow_stats fs;
2770 minimatch_expand(&rule->cr.match, &fs.match);
2771 fs.priority = rule->cr.priority;
2772 fs.cookie = rule->flow_cookie;
2773 fs.table_id = rule->table_id;
2774 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2776 fs.idle_timeout = rule->idle_timeout;
2777 fs.hard_timeout = rule->hard_timeout;
2778 fs.idle_age = age_secs(now - rule->used);
2779 fs.hard_age = age_secs(now - rule->modified);
2780 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2782 fs.ofpacts = rule->ofpacts;
2783 fs.ofpacts_len = rule->ofpacts_len;
2785 if (rule->send_flow_removed) {
2786 fs.flags |= OFPFF_SEND_FLOW_REM;
2787 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2788 and OFPFF13_NO_BYT_COUNTS */
2790 ofputil_append_flow_stats_reply(&fs, &replies);
2792 ofconn_send_replies(ofconn, &replies);
2798 flow_stats_ds(struct rule *rule, struct ds *results)
2800 uint64_t packet_count, byte_count;
2802 rule->ofproto->ofproto_class->rule_get_stats(rule,
2803 &packet_count, &byte_count);
2805 if (rule->table_id != 0) {
2806 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2808 ds_put_format(results, "duration=%llds, ",
2809 (time_msec() - rule->created) / 1000);
2810 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2811 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2812 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2813 cls_rule_format(&rule->cr, results);
2814 ds_put_char(results, ',');
2815 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2816 ds_put_cstr(results, "\n");
2819 /* Adds a pretty-printed description of all flows to 'results', including
2820 * hidden flows (e.g., set up by in-band control). */
2822 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2824 struct oftable *table;
2826 OFPROTO_FOR_EACH_TABLE (table, p) {
2827 struct cls_cursor cursor;
2830 cls_cursor_init(&cursor, &table->cls, NULL);
2831 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2832 flow_stats_ds(rule, results);
2837 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2838 * '*engine_type' and '*engine_id', respectively. */
2840 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2841 uint8_t *engine_type, uint8_t *engine_id)
2843 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2846 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2847 * true if the port's CFM status was successfully stored into '*status'.
2848 * Returns false if the port did not have CFM configured, in which case
2849 * '*status' is indeterminate.
2851 * The caller must provide and owns '*status', but it does not own and must not
2852 * modify or free the array returned in 'status->rmps'. */
2854 ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
2855 struct ofproto_cfm_status *status)
2857 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2859 && ofproto->ofproto_class->get_cfm_status
2860 && ofproto->ofproto_class->get_cfm_status(ofport, status));
2864 handle_aggregate_stats_request(struct ofconn *ofconn,
2865 const struct ofp_header *oh)
2867 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2868 struct ofputil_flow_stats_request request;
2869 struct ofputil_aggregate_stats stats;
2870 bool unknown_packets, unknown_bytes;
2871 struct ofpbuf *reply;
2876 error = ofputil_decode_flow_stats_request(&request, oh);
2881 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2882 request.cookie, request.cookie_mask,
2883 request.out_port, &rules);
2888 memset(&stats, 0, sizeof stats);
2889 unknown_packets = unknown_bytes = false;
2890 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2891 uint64_t packet_count;
2892 uint64_t byte_count;
2894 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2897 if (packet_count == UINT64_MAX) {
2898 unknown_packets = true;
2900 stats.packet_count += packet_count;
2903 if (byte_count == UINT64_MAX) {
2904 unknown_bytes = true;
2906 stats.byte_count += byte_count;
2911 if (unknown_packets) {
2912 stats.packet_count = UINT64_MAX;
2914 if (unknown_bytes) {
2915 stats.byte_count = UINT64_MAX;
2918 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2919 ofconn_send_reply(ofconn, reply);
2924 struct queue_stats_cbdata {
2925 struct ofport *ofport;
2926 struct list replies;
2930 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2931 const struct netdev_queue_stats *stats)
2934 struct ofputil_queue_stats oqs = {
2935 .port_no = cbdata->ofport->pp.port_no,
2936 .queue_id = queue_id,
2939 ofputil_append_queue_stat(&cbdata->replies, &oqs);
2943 handle_queue_stats_dump_cb(uint32_t queue_id,
2944 struct netdev_queue_stats *stats,
2947 struct queue_stats_cbdata *cbdata = cbdata_;
2949 put_queue_stats(cbdata, queue_id, stats);
2953 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2954 struct queue_stats_cbdata *cbdata)
2956 cbdata->ofport = port;
2957 if (queue_id == OFPQ_ALL) {
2958 netdev_dump_queue_stats(port->netdev,
2959 handle_queue_stats_dump_cb, cbdata);
2961 struct netdev_queue_stats stats;
2963 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2964 put_queue_stats(cbdata, queue_id, &stats);
2966 return OFPERR_OFPQOFC_BAD_QUEUE;
2973 handle_queue_stats_request(struct ofconn *ofconn,
2974 const struct ofp_header *rq)
2976 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2977 struct queue_stats_cbdata cbdata;
2978 struct ofport *port;
2980 struct ofputil_queue_stats_request oqsr;
2982 COVERAGE_INC(ofproto_queue_req);
2984 ofpmp_init(&cbdata.replies, rq);
2986 error = ofputil_decode_queue_stats_request(rq, &oqsr);
2991 if (oqsr.port_no == OFPP_ANY) {
2992 error = OFPERR_OFPQOFC_BAD_QUEUE;
2993 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2994 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
2999 port = ofproto_get_port(ofproto, oqsr.port_no);
3001 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3002 : OFPERR_OFPQOFC_BAD_PORT);
3005 ofconn_send_replies(ofconn, &cbdata.replies);
3007 ofpbuf_list_delete(&cbdata.replies);
3014 is_flow_deletion_pending(const struct ofproto *ofproto,
3015 const struct cls_rule *cls_rule,
3018 if (!hmap_is_empty(&ofproto->deletions)) {
3019 struct ofoperation *op;
3021 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3022 cls_rule_hash(cls_rule, table_id),
3023 &ofproto->deletions) {
3024 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3033 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3034 * in which no matching flow already exists in the flow table.
3036 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3037 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3038 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3039 * initiated now but may be retried later.
3041 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3042 * ownership remains with the caller.
3044 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3047 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3048 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3050 struct oftable *table;
3051 struct ofopgroup *group;
3052 struct rule *victim;
3057 error = check_table_id(ofproto, fm->table_id);
3063 if (fm->table_id == 0xff) {
3065 if (ofproto->ofproto_class->rule_choose_table) {
3066 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3072 ovs_assert(table_id < ofproto->n_tables);
3073 table = &ofproto->tables[table_id];
3075 table = &ofproto->tables[0];
3077 } else if (fm->table_id < ofproto->n_tables) {
3078 table = &ofproto->tables[fm->table_id];
3080 return OFPERR_OFPBRC_BAD_TABLE_ID;
3083 if (table->flags & OFTABLE_READONLY) {
3084 return OFPERR_OFPBRC_EPERM;
3087 /* Allocate new rule and initialize classifier rule. */
3088 rule = ofproto->ofproto_class->rule_alloc();
3090 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3091 ofproto->name, strerror(error));
3094 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3096 /* Serialize against pending deletion. */
3097 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3098 cls_rule_destroy(&rule->cr);
3099 ofproto->ofproto_class->rule_dealloc(rule);
3100 return OFPROTO_POSTPONE;
3103 /* Check for overlap, if requested. */
3104 if (fm->flags & OFPFF_CHECK_OVERLAP
3105 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3106 cls_rule_destroy(&rule->cr);
3107 ofproto->ofproto_class->rule_dealloc(rule);
3108 return OFPERR_OFPFMFC_OVERLAP;
3111 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3113 rule->ofproto = ofproto;
3114 rule->pending = NULL;
3115 rule->flow_cookie = fm->new_cookie;
3116 rule->created = rule->modified = rule->used = time_msec();
3117 rule->idle_timeout = fm->idle_timeout;
3118 rule->hard_timeout = fm->hard_timeout;
3119 rule->table_id = table - ofproto->tables;
3120 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3121 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3122 and OFPFF13_NO_BYT_COUNTS */
3123 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3124 rule->ofpacts_len = fm->ofpacts_len;
3125 rule->evictable = true;
3126 rule->eviction_group = NULL;
3127 list_init(&rule->expirable);
3128 rule->monitor_flags = 0;
3129 rule->add_seqno = 0;
3130 rule->modify_seqno = 0;
3132 /* Insert new rule. */
3133 victim = oftable_replace_rule(rule);
3134 if (victim && !rule_is_modifiable(victim)) {
3135 error = OFPERR_OFPBRC_EPERM;
3136 } else if (victim && victim->pending) {
3137 error = OFPROTO_POSTPONE;
3139 struct ofoperation *op;
3142 if (classifier_count(&table->cls) > table->max_flows) {
3145 was_evictable = rule->evictable;
3146 rule->evictable = false;
3147 evict = choose_rule_to_evict(table);
3148 rule->evictable = was_evictable;
3151 error = OFPERR_OFPFMFC_TABLE_FULL;
3153 } else if (evict->pending) {
3154 error = OFPROTO_POSTPONE;
3161 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3162 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3163 op->victim = victim;
3165 error = ofproto->ofproto_class->rule_construct(rule);
3167 op->group->n_running--;
3168 ofoperation_destroy(rule->pending);
3170 delete_flow__(evict, group);
3172 ofopgroup_submit(group);
3176 /* Back out if an error occurred. */
3178 oftable_substitute_rule(rule, victim);
3179 ofproto_rule_destroy__(rule);
3184 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3186 /* Modifies the rules listed in 'rules', changing their actions to match those
3189 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3192 * Returns 0 on success, otherwise an OpenFlow error code. */
3194 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3195 const struct ofputil_flow_mod *fm,
3196 const struct ofp_header *request, struct list *rules)
3198 struct ofopgroup *group;
3202 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3203 error = OFPERR_OFPBRC_EPERM;
3204 LIST_FOR_EACH (rule, ofproto_node, rules) {
3205 struct ofoperation *op;
3206 bool actions_changed;
3207 ovs_be64 new_cookie;
3209 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3211 if (rule_is_modifiable(rule)) {
3212 /* At least one rule is modifiable, don't report EPERM error. */
3218 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3219 rule->ofpacts, rule->ofpacts_len);
3220 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3222 : rule->flow_cookie);
3224 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3225 rule->flow_cookie = new_cookie;
3226 if (actions_changed) {
3227 op->ofpacts = rule->ofpacts;
3228 op->ofpacts_len = rule->ofpacts_len;
3229 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3230 rule->ofpacts_len = fm->ofpacts_len;
3231 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3233 ofoperation_complete(op, 0);
3236 ofopgroup_submit(group);
3242 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3243 const struct ofputil_flow_mod *fm,
3244 const struct ofp_header *request)
3246 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3249 return add_flow(ofproto, ofconn, fm, request);
3252 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3255 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3258 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3259 const struct ofputil_flow_mod *fm,
3260 const struct ofp_header *request)
3265 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3266 fm->cookie, fm->cookie_mask,
3270 } else if (list_is_empty(&rules)) {
3271 return modify_flows_add(ofproto, ofconn, fm, request);
3273 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3277 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3280 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3283 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3284 const struct ofputil_flow_mod *fm,
3285 const struct ofp_header *request)
3290 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3291 fm->priority, fm->cookie, fm->cookie_mask,
3296 } else if (list_is_empty(&rules)) {
3297 return modify_flows_add(ofproto, ofconn, fm, request);
3299 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3300 fm, request, &rules)
3305 /* OFPFC_DELETE implementation. */
3308 delete_flow__(struct rule *rule, struct ofopgroup *group)
3310 struct ofproto *ofproto = rule->ofproto;
3312 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3314 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3315 oftable_remove_rule(rule);
3316 ofproto->ofproto_class->rule_destruct(rule);
3319 /* Deletes the rules listed in 'rules'.
3321 * Returns 0 on success, otherwise an OpenFlow error code. */
3323 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3324 const struct ofp_header *request, struct list *rules)
3326 struct rule *rule, *next;
3327 struct ofopgroup *group;
3329 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3330 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3331 delete_flow__(rule, group);
3333 ofopgroup_submit(group);
3338 /* Implements OFPFC_DELETE. */
3340 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3341 const struct ofputil_flow_mod *fm,
3342 const struct ofp_header *request)
3347 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3348 fm->cookie, fm->cookie_mask,
3349 fm->out_port, &rules);
3350 return (error ? error
3351 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3356 /* Implements OFPFC_DELETE_STRICT. */
3358 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3359 const struct ofputil_flow_mod *fm,
3360 const struct ofp_header *request)
3365 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3366 fm->priority, fm->cookie, fm->cookie_mask,
3367 fm->out_port, &rules);
3368 return (error ? error
3369 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3375 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3377 struct ofputil_flow_removed fr;
3379 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3383 minimatch_expand(&rule->cr.match, &fr.match);
3384 fr.priority = rule->cr.priority;
3385 fr.cookie = rule->flow_cookie;
3387 fr.table_id = rule->table_id;
3388 calc_flow_duration__(rule->created, time_msec(),
3389 &fr.duration_sec, &fr.duration_nsec);
3390 fr.idle_timeout = rule->idle_timeout;
3391 fr.hard_timeout = rule->hard_timeout;
3392 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3395 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3399 ofproto_rule_update_used(struct rule *rule, long long int used)
3401 if (used > rule->used) {
3402 struct eviction_group *evg = rule->eviction_group;
3406 heap_change(&evg->rules, &rule->evg_node,
3407 rule_eviction_priority(rule));
3412 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3413 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3416 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3419 * ofproto implementation ->run() functions should use this function to expire
3420 * OpenFlow flows. */
3422 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3424 struct ofproto *ofproto = rule->ofproto;
3425 struct ofopgroup *group;
3427 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3429 ofproto_rule_send_removed(rule, reason);
3431 group = ofopgroup_create_unattached(ofproto);
3432 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3433 oftable_remove_rule(rule);
3434 ofproto->ofproto_class->rule_destruct(rule);
3435 ofopgroup_submit(group);
3439 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3441 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3442 struct ofputil_flow_mod fm;
3443 uint64_t ofpacts_stub[1024 / 8];
3444 struct ofpbuf ofpacts;
3448 error = reject_slave_controller(ofconn);
3453 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3454 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3457 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3458 &fm.match.flow, ofproto->max_ports);
3461 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3464 goto exit_free_ofpacts;
3467 /* Record the operation for logging a summary report. */
3468 switch (fm.command) {
3474 case OFPFC_MODIFY_STRICT:
3475 ofproto->n_modify++;
3479 case OFPFC_DELETE_STRICT:
3480 ofproto->n_delete++;
3485 if (ofproto->next_op_report == LLONG_MAX) {
3486 ofproto->first_op = now;
3487 ofproto->next_op_report = MAX(now + 10 * 1000,
3488 ofproto->op_backoff);
3489 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3491 ofproto->last_op = now;
3494 ofpbuf_uninit(&ofpacts);
3500 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3501 const struct ofputil_flow_mod *fm,
3502 const struct ofp_header *oh)
3504 if (ofproto->n_pending >= 50) {
3505 ovs_assert(!list_is_empty(&ofproto->pending));
3506 return OFPROTO_POSTPONE;
3509 switch (fm->command) {
3511 return add_flow(ofproto, ofconn, fm, oh);
3514 return modify_flows_loose(ofproto, ofconn, fm, oh);
3516 case OFPFC_MODIFY_STRICT:
3517 return modify_flow_strict(ofproto, ofconn, fm, oh);
3520 return delete_flows_loose(ofproto, ofconn, fm, oh);
3522 case OFPFC_DELETE_STRICT:
3523 return delete_flow_strict(ofproto, ofconn, fm, oh);
3526 if (fm->command > 0xff) {
3527 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3528 "flow_mod_table_id extension is not enabled",
3531 return OFPERR_OFPFMFC_BAD_COMMAND;
3536 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3538 struct ofputil_role_request request;
3539 struct ofputil_role_request reply;
3543 error = ofputil_decode_role_message(oh, &request);
3548 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3549 if (ofconn_get_role(ofconn) != request.role
3550 && ofconn_has_pending_opgroups(ofconn)) {
3551 return OFPROTO_POSTPONE;
3554 if (request.have_generation_id
3555 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3556 return OFPERR_OFPRRFC_STALE;
3559 ofconn_set_role(ofconn, request.role);
3562 reply.role = ofconn_get_role(ofconn);
3563 reply.have_generation_id = ofconn_get_master_election_id(
3564 ofconn, &reply.generation_id);
3565 buf = ofputil_encode_role_reply(oh, &reply);
3566 ofconn_send_reply(ofconn, buf);
3572 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3573 const struct ofp_header *oh)
3575 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3576 enum ofputil_protocol cur, next;
3578 cur = ofconn_get_protocol(ofconn);
3579 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3580 ofconn_set_protocol(ofconn, next);
3586 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3588 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3589 enum ofputil_protocol cur, next;
3590 enum ofputil_protocol next_base;
3592 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3594 return OFPERR_OFPBRC_EPERM;
3597 cur = ofconn_get_protocol(ofconn);
3598 next = ofputil_protocol_set_base(cur, next_base);
3599 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3600 /* Avoid sending async messages in surprising protocol. */
3601 return OFPROTO_POSTPONE;
3604 ofconn_set_protocol(ofconn, next);
3609 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3610 const struct ofp_header *oh)
3612 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3615 format = ntohl(msg->format);
3616 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3617 return OFPERR_OFPBRC_EPERM;
3620 if (format != ofconn_get_packet_in_format(ofconn)
3621 && ofconn_has_pending_opgroups(ofconn)) {
3622 /* Avoid sending async message in surprsing packet in format. */
3623 return OFPROTO_POSTPONE;
3626 ofconn_set_packet_in_format(ofconn, format);
3631 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3633 const struct nx_async_config *msg = ofpmsg_body(oh);
3634 uint32_t master[OAM_N_TYPES];
3635 uint32_t slave[OAM_N_TYPES];
3637 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3638 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3639 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3641 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3642 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3643 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3645 ofconn_set_async_config(ofconn, master, slave);
3646 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3647 !ofconn_get_miss_send_len(ofconn)) {
3648 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3655 handle_nxt_set_controller_id(struct ofconn *ofconn,
3656 const struct ofp_header *oh)
3658 const struct nx_controller_id *nci = ofpmsg_body(oh);
3660 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3661 return OFPERR_NXBRC_MUST_BE_ZERO;
3664 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3669 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3673 if (ofconn_has_pending_opgroups(ofconn)) {
3674 return OFPROTO_POSTPONE;
3677 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3678 ? OFPRAW_OFPT10_BARRIER_REPLY
3679 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3680 ofconn_send_reply(ofconn, buf);
3685 ofproto_compose_flow_refresh_update(const struct rule *rule,
3686 enum nx_flow_monitor_flags flags,
3689 struct ofoperation *op = rule->pending;
3690 struct ofputil_flow_update fu;
3693 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3694 /* We'll report the final flow when the operation completes. Reporting
3695 * it now would cause a duplicate report later. */
3699 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3700 ? NXFME_ADDED : NXFME_MODIFIED);
3702 fu.idle_timeout = rule->idle_timeout;
3703 fu.hard_timeout = rule->hard_timeout;
3704 fu.table_id = rule->table_id;
3705 fu.cookie = rule->flow_cookie;
3706 minimatch_expand(&rule->cr.match, &match);
3708 fu.priority = rule->cr.priority;
3709 if (!(flags & NXFMF_ACTIONS)) {
3713 fu.ofpacts = rule->ofpacts;
3714 fu.ofpacts_len = rule->ofpacts_len;
3716 /* An operation is in progress. Use the previous version of the flow's
3717 * actions, so that when the operation commits we report the change. */
3719 case OFOPERATION_ADD:
3720 /* We already verified that there was a victim. */
3721 fu.ofpacts = op->victim->ofpacts;
3722 fu.ofpacts_len = op->victim->ofpacts_len;
3725 case OFOPERATION_MODIFY:
3727 fu.ofpacts = op->ofpacts;
3728 fu.ofpacts_len = op->ofpacts_len;
3730 fu.ofpacts = rule->ofpacts;
3731 fu.ofpacts_len = rule->ofpacts_len;
3735 case OFOPERATION_DELETE:
3736 fu.ofpacts = rule->ofpacts;
3737 fu.ofpacts_len = rule->ofpacts_len;
3745 if (list_is_empty(msgs)) {
3746 ofputil_start_flow_update(msgs);
3748 ofputil_append_flow_update(&fu, msgs);
3752 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3756 LIST_FOR_EACH (rule, ofproto_node, rules) {
3757 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3758 rule->monitor_flags = 0;
3760 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3765 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3766 struct rule *rule, uint64_t seqno,
3769 enum nx_flow_monitor_flags update;
3771 if (ofproto_rule_is_hidden(rule)) {
3776 ? ofoperation_has_out_port(rule->pending, m->out_port)
3777 : ofproto_rule_has_out_port(rule, m->out_port))) {
3782 if (rule->add_seqno > seqno) {
3783 update = NXFMF_ADD | NXFMF_MODIFY;
3784 } else if (rule->modify_seqno > seqno) {
3785 update = NXFMF_MODIFY;
3790 if (!(m->flags & update)) {
3794 update = NXFMF_INITIAL;
3797 if (!rule->monitor_flags) {
3798 list_push_back(rules, &rule->ofproto_node);
3800 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3804 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3808 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3809 const struct ofoperation *op;
3810 const struct oftable *table;
3811 struct cls_rule target;
3813 cls_rule_init_from_minimatch(&target, &m->match, 0);
3814 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3815 struct cls_cursor cursor;
3818 cls_cursor_init(&cursor, &table->cls, &target);
3819 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3820 ovs_assert(!rule->pending); /* XXX */
3821 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3825 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3826 struct rule *rule = op->rule;
3828 if (((m->table_id == 0xff
3829 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3830 : m->table_id == rule->table_id))
3831 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3832 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3835 cls_rule_destroy(&target);
3839 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3842 if (m->flags & NXFMF_INITIAL) {
3843 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3848 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3849 uint64_t seqno, struct list *rules)
3851 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3855 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3857 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3858 struct ofmonitor **monitors;
3859 size_t n_monitors, allocated_monitors;
3860 struct list replies;
3867 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3869 n_monitors = allocated_monitors = 0;
3871 struct ofputil_flow_monitor_request request;
3872 struct ofmonitor *m;
3875 retval = ofputil_decode_flow_monitor_request(&request, &b);
3876 if (retval == EOF) {
3878 } else if (retval) {
3883 if (request.table_id != 0xff
3884 && request.table_id >= ofproto->n_tables) {
3885 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3889 error = ofmonitor_create(&request, ofconn, &m);
3894 if (n_monitors >= allocated_monitors) {
3895 monitors = x2nrealloc(monitors, &allocated_monitors,
3898 monitors[n_monitors++] = m;
3902 for (i = 0; i < n_monitors; i++) {
3903 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3906 ofpmp_init(&replies, oh);
3907 ofmonitor_compose_refresh_updates(&rules, &replies);
3908 ofconn_send_replies(ofconn, &replies);
3915 for (i = 0; i < n_monitors; i++) {
3916 ofmonitor_destroy(monitors[i]);
3923 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3925 struct ofmonitor *m;
3928 id = ofputil_decode_flow_monitor_cancel(oh);
3929 m = ofmonitor_lookup(ofconn, id);
3931 return OFPERR_NXBRC_FM_BAD_ID;
3934 ofmonitor_destroy(m);
3939 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3941 const struct ofp_header *oh = msg->data;
3945 error = ofptype_decode(&type, oh);
3951 /* OpenFlow requests. */
3952 case OFPTYPE_ECHO_REQUEST:
3953 return handle_echo_request(ofconn, oh);
3955 case OFPTYPE_FEATURES_REQUEST:
3956 return handle_features_request(ofconn, oh);
3958 case OFPTYPE_GET_CONFIG_REQUEST:
3959 return handle_get_config_request(ofconn, oh);
3961 case OFPTYPE_SET_CONFIG:
3962 return handle_set_config(ofconn, oh);
3964 case OFPTYPE_PACKET_OUT:
3965 return handle_packet_out(ofconn, oh);
3967 case OFPTYPE_PORT_MOD:
3968 return handle_port_mod(ofconn, oh);
3970 case OFPTYPE_FLOW_MOD:
3971 return handle_flow_mod(ofconn, oh);
3973 case OFPTYPE_BARRIER_REQUEST:
3974 return handle_barrier_request(ofconn, oh);
3976 case OFPTYPE_ROLE_REQUEST:
3977 return handle_role_request(ofconn, oh);
3979 /* OpenFlow replies. */
3980 case OFPTYPE_ECHO_REPLY:
3983 /* Nicira extension requests. */
3984 case OFPTYPE_FLOW_MOD_TABLE_ID:
3985 return handle_nxt_flow_mod_table_id(ofconn, oh);
3987 case OFPTYPE_SET_FLOW_FORMAT:
3988 return handle_nxt_set_flow_format(ofconn, oh);
3990 case OFPTYPE_SET_PACKET_IN_FORMAT:
3991 return handle_nxt_set_packet_in_format(ofconn, oh);
3993 case OFPTYPE_SET_CONTROLLER_ID:
3994 return handle_nxt_set_controller_id(ofconn, oh);
3996 case OFPTYPE_FLOW_AGE:
3997 /* Nothing to do. */
4000 case OFPTYPE_FLOW_MONITOR_CANCEL:
4001 return handle_flow_monitor_cancel(ofconn, oh);
4003 case OFPTYPE_SET_ASYNC_CONFIG:
4004 return handle_nxt_set_async_config(ofconn, oh);
4006 /* Statistics requests. */
4007 case OFPTYPE_DESC_STATS_REQUEST:
4008 return handle_desc_stats_request(ofconn, oh);
4010 case OFPTYPE_FLOW_STATS_REQUEST:
4011 return handle_flow_stats_request(ofconn, oh);
4013 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4014 return handle_aggregate_stats_request(ofconn, oh);
4016 case OFPTYPE_TABLE_STATS_REQUEST:
4017 return handle_table_stats_request(ofconn, oh);
4019 case OFPTYPE_PORT_STATS_REQUEST:
4020 return handle_port_stats_request(ofconn, oh);
4022 case OFPTYPE_QUEUE_STATS_REQUEST:
4023 return handle_queue_stats_request(ofconn, oh);
4025 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4026 return handle_port_desc_stats_request(ofconn, oh);
4028 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4029 return handle_flow_monitor_request(ofconn, oh);
4031 /* FIXME: Change the following once they are implemented: */
4032 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4033 case OFPTYPE_GET_ASYNC_REQUEST:
4034 case OFPTYPE_METER_MOD:
4035 case OFPTYPE_GROUP_REQUEST:
4036 case OFPTYPE_GROUP_DESC_REQUEST:
4037 case OFPTYPE_GROUP_FEATURES_REQUEST:
4038 case OFPTYPE_METER_REQUEST:
4039 case OFPTYPE_METER_CONFIG_REQUEST:
4040 case OFPTYPE_METER_FEATURES_REQUEST:
4041 case OFPTYPE_TABLE_FEATURES_REQUEST:
4042 return OFPERR_OFPBRC_BAD_TYPE;
4046 case OFPTYPE_FEATURES_REPLY:
4047 case OFPTYPE_GET_CONFIG_REPLY:
4048 case OFPTYPE_PACKET_IN:
4049 case OFPTYPE_FLOW_REMOVED:
4050 case OFPTYPE_PORT_STATUS:
4051 case OFPTYPE_BARRIER_REPLY:
4052 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4053 case OFPTYPE_DESC_STATS_REPLY:
4054 case OFPTYPE_FLOW_STATS_REPLY:
4055 case OFPTYPE_QUEUE_STATS_REPLY:
4056 case OFPTYPE_PORT_STATS_REPLY:
4057 case OFPTYPE_TABLE_STATS_REPLY:
4058 case OFPTYPE_AGGREGATE_STATS_REPLY:
4059 case OFPTYPE_PORT_DESC_STATS_REPLY:
4060 case OFPTYPE_ROLE_REPLY:
4061 case OFPTYPE_FLOW_MONITOR_PAUSED:
4062 case OFPTYPE_FLOW_MONITOR_RESUMED:
4063 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4064 case OFPTYPE_GET_ASYNC_REPLY:
4065 case OFPTYPE_GROUP_REPLY:
4066 case OFPTYPE_GROUP_DESC_REPLY:
4067 case OFPTYPE_GROUP_FEATURES_REPLY:
4068 case OFPTYPE_METER_REPLY:
4069 case OFPTYPE_METER_CONFIG_REPLY:
4070 case OFPTYPE_METER_FEATURES_REPLY:
4071 case OFPTYPE_TABLE_FEATURES_REPLY:
4073 return OFPERR_OFPBRC_BAD_TYPE;
4078 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4080 int error = handle_openflow__(ofconn, ofp_msg);
4081 if (error && error != OFPROTO_POSTPONE) {
4082 ofconn_send_error(ofconn, ofp_msg->data, error);
4084 COVERAGE_INC(ofproto_recv_openflow);
4085 return error != OFPROTO_POSTPONE;
4088 /* Asynchronous operations. */
4090 /* Creates and returns a new ofopgroup that is not associated with any
4091 * OpenFlow connection.
4093 * The caller should add operations to the returned group with
4094 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4095 static struct ofopgroup *
4096 ofopgroup_create_unattached(struct ofproto *ofproto)
4098 struct ofopgroup *group = xzalloc(sizeof *group);
4099 group->ofproto = ofproto;
4100 list_init(&group->ofproto_node);
4101 list_init(&group->ops);
4102 list_init(&group->ofconn_node);
4106 /* Creates and returns a new ofopgroup for 'ofproto'.
4108 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4109 * connection. The 'request' and 'buffer_id' arguments are ignored.
4111 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4112 * If the ofopgroup eventually fails, then the error reply will include
4113 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4114 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4116 * The caller should add operations to the returned group with
4117 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4118 static struct ofopgroup *
4119 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4120 const struct ofp_header *request, uint32_t buffer_id)
4122 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4124 size_t request_len = ntohs(request->length);
4126 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4128 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4129 group->ofconn = ofconn;
4130 group->request = xmemdup(request, MIN(request_len, 64));
4131 group->buffer_id = buffer_id;
4136 /* Submits 'group' for processing.
4138 * If 'group' contains no operations (e.g. none were ever added, or all of the
4139 * ones that were added completed synchronously), then it is destroyed
4140 * immediately. Otherwise it is added to the ofproto's list of pending
4143 ofopgroup_submit(struct ofopgroup *group)
4145 if (!group->n_running) {
4146 ofopgroup_complete(group);
4148 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4149 group->ofproto->n_pending++;
4154 ofopgroup_complete(struct ofopgroup *group)
4156 struct ofproto *ofproto = group->ofproto;
4158 struct ofconn *abbrev_ofconn;
4159 ovs_be32 abbrev_xid;
4161 struct ofoperation *op, *next_op;
4164 ovs_assert(!group->n_running);
4167 LIST_FOR_EACH (op, group_node, &group->ops) {
4174 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4175 LIST_FOR_EACH (op, group_node, &group->ops) {
4176 if (op->type != OFOPERATION_DELETE) {
4177 struct ofpbuf *packet;
4180 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4184 error = rule_execute(op->rule, in_port, packet);
4191 if (!error && !list_is_empty(&group->ofconn_node)) {
4192 abbrev_ofconn = group->ofconn;
4193 abbrev_xid = group->request->xid;
4195 abbrev_ofconn = NULL;
4196 abbrev_xid = htonl(0);
4198 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4199 struct rule *rule = op->rule;
4201 /* We generally want to report the change to active OpenFlow flow
4202 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4204 - The operation failed.
4206 - The affected rule is not visible to controllers.
4208 - The operation's only effect was to update rule->modified. */
4210 || ofproto_rule_is_hidden(rule)
4211 || (op->type == OFOPERATION_MODIFY
4213 && rule->flow_cookie == op->flow_cookie))) {
4214 /* Check that we can just cast from ofoperation_type to
4215 * nx_flow_update_event. */
4216 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4218 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4220 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4223 ofmonitor_report(ofproto->connmgr, rule,
4224 (enum nx_flow_update_event) op->type,
4225 op->reason, abbrev_ofconn, abbrev_xid);
4228 rule->pending = NULL;
4231 case OFOPERATION_ADD:
4235 ofproto_rule_destroy__(op->victim);
4236 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4237 if (vid_mask == VLAN_VID_MASK) {
4238 if (ofproto->vlan_bitmap) {
4239 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4240 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4241 bitmap_set1(ofproto->vlan_bitmap, vid);
4242 ofproto->vlans_changed = true;
4245 ofproto->vlans_changed = true;
4249 oftable_substitute_rule(rule, op->victim);
4250 ofproto_rule_destroy__(rule);
4254 case OFOPERATION_DELETE:
4255 ovs_assert(!op->error);
4256 ofproto_rule_destroy__(rule);
4260 case OFOPERATION_MODIFY:
4262 rule->modified = time_msec();
4264 rule->flow_cookie = op->flow_cookie;
4266 free(rule->ofpacts);
4267 rule->ofpacts = op->ofpacts;
4268 rule->ofpacts_len = op->ofpacts_len;
4270 op->ofpacts_len = 0;
4279 ofoperation_destroy(op);
4282 ofmonitor_flush(ofproto->connmgr);
4284 if (!list_is_empty(&group->ofproto_node)) {
4285 ovs_assert(ofproto->n_pending > 0);
4286 ofproto->n_pending--;
4287 list_remove(&group->ofproto_node);
4289 if (!list_is_empty(&group->ofconn_node)) {
4290 list_remove(&group->ofconn_node);
4292 ofconn_send_error(group->ofconn, group->request, error);
4294 connmgr_retry(ofproto->connmgr);
4296 free(group->request);
4300 /* Initiates a new operation on 'rule', of the specified 'type', within
4301 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4303 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4304 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4306 * Returns the newly created ofoperation (which is also available as
4307 * rule->pending). */
4308 static struct ofoperation *
4309 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4310 enum ofoperation_type type,
4311 enum ofp_flow_removed_reason reason)
4313 struct ofproto *ofproto = group->ofproto;
4314 struct ofoperation *op;
4316 ovs_assert(!rule->pending);
4318 op = rule->pending = xzalloc(sizeof *op);
4320 list_push_back(&group->ops, &op->group_node);
4323 op->reason = reason;
4324 op->flow_cookie = rule->flow_cookie;
4328 if (type == OFOPERATION_DELETE) {
4329 hmap_insert(&ofproto->deletions, &op->hmap_node,
4330 cls_rule_hash(&rule->cr, rule->table_id));
4337 ofoperation_destroy(struct ofoperation *op)
4339 struct ofopgroup *group = op->group;
4342 op->rule->pending = NULL;
4344 if (op->type == OFOPERATION_DELETE) {
4345 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4347 list_remove(&op->group_node);
4352 /* Indicates that 'op' completed with status 'error', which is either 0 to
4353 * indicate success or an OpenFlow error code on failure.
4355 * If 'error' is 0, indicating success, the operation will be committed
4356 * permanently to the flow table. There is one interesting subcase:
4358 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4359 * the flow table (the "victim" rule) by a new one, then the caller must
4360 * have uninitialized any derived state in the victim rule, as in step 5 in
4361 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4362 * performs steps 6 and 7 for the victim rule, most notably by calling its
4363 * ->rule_dealloc() function.
4365 * If 'error' is nonzero, then generally the operation will be rolled back:
4367 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4368 * restores the original rule. The caller must have uninitialized any
4369 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4370 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4371 * and 7 for the new rule, calling its ->rule_dealloc() function.
4373 * - If 'op' is a "modify flow" operation, ofproto restores the original
4376 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4377 * allowed to fail. It must always succeed.
4379 * Please see the large comment in ofproto/ofproto-provider.h titled
4380 * "Asynchronous Operation Support" for more information. */
4382 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4384 struct ofopgroup *group = op->group;
4386 ovs_assert(op->rule->pending == op);
4387 ovs_assert(group->n_running > 0);
4388 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4391 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4392 ofopgroup_complete(group);
4397 ofoperation_get_victim(struct ofoperation *op)
4399 ovs_assert(op->type == OFOPERATION_ADD);
4404 pick_datapath_id(const struct ofproto *ofproto)
4406 const struct ofport *port;
4408 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4410 uint8_t ea[ETH_ADDR_LEN];
4413 error = netdev_get_etheraddr(port->netdev, ea);
4415 return eth_addr_to_uint64(ea);
4417 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4418 ofproto->name, netdev_get_name(port->netdev),
4421 return ofproto->fallback_dpid;
4425 pick_fallback_dpid(void)
4427 uint8_t ea[ETH_ADDR_LEN];
4428 eth_addr_nicira_random(ea);
4429 return eth_addr_to_uint64(ea);
4432 /* Table overflow policy. */
4434 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4435 * is not configured to evict rules or if the table contains no evictable
4436 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4438 static struct rule *
4439 choose_rule_to_evict(struct oftable *table)
4441 struct eviction_group *evg;
4443 if (!table->eviction_fields) {
4447 /* In the common case, the outer and inner loops here will each be entered
4450 * - The inner loop normally "return"s in its first iteration. If the
4451 * eviction group has any evictable rules, then it always returns in
4454 * - The outer loop only iterates more than once if the largest eviction
4455 * group has no evictable rules.
4457 * - The outer loop can exit only if table's 'max_flows' is all filled up
4458 * by unevictable rules'. */
4459 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4462 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4463 if (rule->evictable) {
4472 /* Searches 'ofproto' for tables that have more flows than their configured
4473 * maximum and that have flow eviction enabled, and evicts as many flows as
4474 * necessary and currently feasible from them.
4476 * This triggers only when an OpenFlow table has N flows in it and then the
4477 * client configures a maximum number of flows less than N. */
4479 ofproto_evict(struct ofproto *ofproto)
4481 struct ofopgroup *group;
4482 struct oftable *table;
4484 group = ofopgroup_create_unattached(ofproto);
4485 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4486 while (classifier_count(&table->cls) > table->max_flows
4487 && table->eviction_fields) {
4490 rule = choose_rule_to_evict(table);
4491 if (!rule || rule->pending) {
4495 ofoperation_create(group, rule,
4496 OFOPERATION_DELETE, OFPRR_EVICTION);
4497 oftable_remove_rule(rule);
4498 ofproto->ofproto_class->rule_destruct(rule);
4501 ofopgroup_submit(group);
4504 /* Eviction groups. */
4506 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4507 * rules. The priority contains low-order random bits to ensure that eviction
4508 * groups with the same number of rules are prioritized randomly. */
4510 eviction_group_priority(size_t n_rules)
4512 uint16_t size = MIN(UINT16_MAX, n_rules);
4513 return (size << 16) | random_uint16();
4516 /* Updates 'evg', an eviction_group within 'table', following a change that
4517 * adds or removes rules in 'evg'. */
4519 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4521 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4522 eviction_group_priority(heap_count(&evg->rules)));
4525 /* Destroys 'evg', an eviction_group within 'table':
4527 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4528 * rules themselves, just removes them from the eviction group.)
4530 * - Removes 'evg' from 'table'.
4534 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4536 while (!heap_is_empty(&evg->rules)) {
4539 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4540 rule->eviction_group = NULL;
4542 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4543 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4544 heap_destroy(&evg->rules);
4548 /* Removes 'rule' from its eviction group, if any. */
4550 eviction_group_remove_rule(struct rule *rule)
4552 if (rule->eviction_group) {
4553 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4554 struct eviction_group *evg = rule->eviction_group;
4556 rule->eviction_group = NULL;
4557 heap_remove(&evg->rules, &rule->evg_node);
4558 if (heap_is_empty(&evg->rules)) {
4559 eviction_group_destroy(table, evg);
4561 eviction_group_resized(table, evg);
4566 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4567 * returns the hash value. */
4569 eviction_group_hash_rule(struct rule *rule)
4571 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4572 const struct mf_subfield *sf;
4576 hash = table->eviction_group_id_basis;
4577 miniflow_expand(&rule->cr.match.flow, &flow);
4578 for (sf = table->eviction_fields;
4579 sf < &table->eviction_fields[table->n_eviction_fields];
4582 if (mf_are_prereqs_ok(sf->field, &flow)) {
4583 union mf_value value;
4585 mf_get_value(sf->field, &flow, &value);
4587 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4589 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4590 unsigned int start = sf->ofs + sf->n_bits;
4591 bitwise_zero(&value, sf->field->n_bytes, start,
4592 sf->field->n_bytes * 8 - start);
4594 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4596 hash = hash_int(hash, 0);
4603 /* Returns an eviction group within 'table' with the given 'id', creating one
4605 static struct eviction_group *
4606 eviction_group_find(struct oftable *table, uint32_t id)
4608 struct eviction_group *evg;
4610 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4614 evg = xmalloc(sizeof *evg);
4615 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4616 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4617 eviction_group_priority(0));
4618 heap_init(&evg->rules);
4623 /* Returns an eviction priority for 'rule'. The return value should be
4624 * interpreted so that higher priorities make a rule more attractive candidates
4627 rule_eviction_priority(struct rule *rule)
4629 long long int hard_expiration;
4630 long long int idle_expiration;
4631 long long int expiration;
4632 uint32_t expiration_offset;
4634 /* Calculate time of expiration. */
4635 hard_expiration = (rule->hard_timeout
4636 ? rule->modified + rule->hard_timeout * 1000
4638 idle_expiration = (rule->idle_timeout
4639 ? rule->used + rule->idle_timeout * 1000
4641 expiration = MIN(hard_expiration, idle_expiration);
4642 if (expiration == LLONG_MAX) {
4646 /* Calculate the time of expiration as a number of (approximate) seconds
4647 * after program startup.
4649 * This should work OK for program runs that last UINT32_MAX seconds or
4650 * less. Therefore, please restart OVS at least once every 136 years. */
4651 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4653 /* Invert the expiration offset because we're using a max-heap. */
4654 return UINT32_MAX - expiration_offset;
4657 /* Adds 'rule' to an appropriate eviction group for its oftable's
4658 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4659 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4662 * The caller must ensure that 'rule' is not already in an eviction group. */
4664 eviction_group_add_rule(struct rule *rule)
4666 struct ofproto *ofproto = rule->ofproto;
4667 struct oftable *table = &ofproto->tables[rule->table_id];
4669 if (table->eviction_fields
4670 && (rule->hard_timeout || rule->idle_timeout)) {
4671 struct eviction_group *evg;
4673 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4675 rule->eviction_group = evg;
4676 heap_insert(&evg->rules, &rule->evg_node,
4677 rule_eviction_priority(rule));
4678 eviction_group_resized(table, evg);
4684 /* Initializes 'table'. */
4686 oftable_init(struct oftable *table)
4688 memset(table, 0, sizeof *table);
4689 classifier_init(&table->cls);
4690 table->max_flows = UINT_MAX;
4693 /* Destroys 'table', including its classifier and eviction groups.
4695 * The caller is responsible for freeing 'table' itself. */
4697 oftable_destroy(struct oftable *table)
4699 ovs_assert(classifier_is_empty(&table->cls));
4700 oftable_disable_eviction(table);
4701 classifier_destroy(&table->cls);
4705 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4706 * string, then 'table' will use its default name.
4708 * This only affects the name exposed for a table exposed through the OpenFlow
4709 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4711 oftable_set_name(struct oftable *table, const char *name)
4713 if (name && name[0]) {
4714 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4715 if (!table->name || strncmp(name, table->name, len)) {
4717 table->name = xmemdup0(name, len);
4725 /* oftables support a choice of two policies when adding a rule would cause the
4726 * number of flows in the table to exceed the configured maximum number: either
4727 * they can refuse to add the new flow or they can evict some existing flow.
4728 * This function configures the former policy on 'table'. */
4730 oftable_disable_eviction(struct oftable *table)
4732 if (table->eviction_fields) {
4733 struct eviction_group *evg, *next;
4735 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4736 &table->eviction_groups_by_id) {
4737 eviction_group_destroy(table, evg);
4739 hmap_destroy(&table->eviction_groups_by_id);
4740 heap_destroy(&table->eviction_groups_by_size);
4742 free(table->eviction_fields);
4743 table->eviction_fields = NULL;
4744 table->n_eviction_fields = 0;
4748 /* oftables support a choice of two policies when adding a rule would cause the
4749 * number of flows in the table to exceed the configured maximum number: either
4750 * they can refuse to add the new flow or they can evict some existing flow.
4751 * This function configures the latter policy on 'table', with fairness based
4752 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4753 * 'n_fields' as 0 disables fairness.) */
4755 oftable_enable_eviction(struct oftable *table,
4756 const struct mf_subfield *fields, size_t n_fields)
4758 struct cls_cursor cursor;
4761 if (table->eviction_fields
4762 && n_fields == table->n_eviction_fields
4764 || !memcmp(fields, table->eviction_fields,
4765 n_fields * sizeof *fields))) {
4770 oftable_disable_eviction(table);
4772 table->n_eviction_fields = n_fields;
4773 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4775 table->eviction_group_id_basis = random_uint32();
4776 hmap_init(&table->eviction_groups_by_id);
4777 heap_init(&table->eviction_groups_by_size);
4779 cls_cursor_init(&cursor, &table->cls, NULL);
4780 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4781 eviction_group_add_rule(rule);
4785 /* Removes 'rule' from the oftable that contains it. */
4787 oftable_remove_rule(struct rule *rule)
4789 struct ofproto *ofproto = rule->ofproto;
4790 struct oftable *table = &ofproto->tables[rule->table_id];
4792 classifier_remove(&table->cls, &rule->cr);
4793 eviction_group_remove_rule(rule);
4794 if (!list_is_empty(&rule->expirable)) {
4795 list_remove(&rule->expirable);
4799 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4800 * oftable that has an identical cls_rule. Returns the rule that was removed,
4801 * if any, and otherwise NULL. */
4802 static struct rule *
4803 oftable_replace_rule(struct rule *rule)
4805 struct ofproto *ofproto = rule->ofproto;
4806 struct oftable *table = &ofproto->tables[rule->table_id];
4807 struct rule *victim;
4808 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4811 list_insert(&ofproto->expirable, &rule->expirable);
4814 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4816 if (!list_is_empty(&victim->expirable)) {
4817 list_remove(&victim->expirable);
4819 eviction_group_remove_rule(victim);
4821 eviction_group_add_rule(rule);
4825 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4827 oftable_substitute_rule(struct rule *old, struct rule *new)
4830 oftable_replace_rule(new);
4832 oftable_remove_rule(old);
4836 /* unixctl commands. */
4839 ofproto_lookup(const char *name)
4841 struct ofproto *ofproto;
4843 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4845 if (!strcmp(ofproto->name, name)) {
4853 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4854 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4856 struct ofproto *ofproto;
4860 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4861 ds_put_format(&results, "%s\n", ofproto->name);
4863 unixctl_command_reply(conn, ds_cstr(&results));
4864 ds_destroy(&results);
4868 ofproto_unixctl_init(void)
4870 static bool registered;
4876 unixctl_command_register("ofproto/list", "", 0, 0,
4877 ofproto_unixctl_list, NULL);
4880 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4882 * This is deprecated. It is only for compatibility with broken device drivers
4883 * in old versions of Linux that do not properly support VLANs when VLAN
4884 * devices are not used. When broken device drivers are no longer in
4885 * widespread use, we will delete these interfaces. */
4887 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4888 * (exactly) by an OpenFlow rule in 'ofproto'. */
4890 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4892 const struct oftable *oftable;
4894 free(ofproto->vlan_bitmap);
4895 ofproto->vlan_bitmap = bitmap_allocate(4096);
4896 ofproto->vlans_changed = false;
4898 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4899 const struct cls_table *table;
4901 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4902 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4903 const struct cls_rule *rule;
4905 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4906 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4907 bitmap_set1(vlan_bitmap, vid);
4908 bitmap_set1(ofproto->vlan_bitmap, vid);
4915 /* Returns true if new VLANs have come into use by the flow table since the
4916 * last call to ofproto_get_vlan_usage().
4918 * We don't track when old VLANs stop being used. */
4920 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4922 return ofproto->vlans_changed;
4925 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4926 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4927 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4928 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4929 * then the VLAN device is un-enslaved. */
4931 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4932 uint16_t realdev_ofp_port, int vid)
4934 struct ofport *ofport;
4937 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
4939 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4941 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4942 ofproto->name, vlandev_ofp_port);
4946 if (!ofproto->ofproto_class->set_realdev) {
4947 if (!vlandev_ofp_port) {
4950 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4954 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4956 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4957 ofproto->name, vlandev_ofp_port,
4958 netdev_get_name(ofport->netdev), strerror(error));