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;
645 ofproto_set_ipfix(struct ofproto *ofproto,
646 const struct ofproto_ipfix_bridge_exporter_options *bo,
647 const struct ofproto_ipfix_flow_exporter_options *fo,
650 if (ofproto->ofproto_class->set_ipfix) {
651 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
653 return (bo || fo) ? EOPNOTSUPP : 0;
657 /* Spanning Tree Protocol (STP) configuration. */
659 /* Configures STP on 'ofproto' using the settings defined in 's'. If
660 * 's' is NULL, disables STP.
662 * Returns 0 if successful, otherwise a positive errno value. */
664 ofproto_set_stp(struct ofproto *ofproto,
665 const struct ofproto_stp_settings *s)
667 return (ofproto->ofproto_class->set_stp
668 ? ofproto->ofproto_class->set_stp(ofproto, s)
672 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
673 * 'enabled' member of 's' is false, then the other members are not
676 * Returns 0 if successful, otherwise a positive errno value. */
678 ofproto_get_stp_status(struct ofproto *ofproto,
679 struct ofproto_stp_status *s)
681 return (ofproto->ofproto_class->get_stp_status
682 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
686 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
687 * in 's'. The caller is responsible for assigning STP port numbers
688 * (using the 'port_num' member in the range of 1 through 255, inclusive)
689 * and ensuring there are no duplicates. If the 's' is NULL, then STP
690 * is disabled on the port.
692 * Returns 0 if successful, otherwise a positive errno value.*/
694 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
695 const struct ofproto_port_stp_settings *s)
697 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
699 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
700 ofproto->name, ofp_port);
704 return (ofproto->ofproto_class->set_stp_port
705 ? ofproto->ofproto_class->set_stp_port(ofport, s)
709 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
710 * 's'. If the 'enabled' member in 's' is false, then the other members
711 * are not meaningful.
713 * Returns 0 if successful, otherwise a positive errno value.*/
715 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
716 struct ofproto_port_stp_status *s)
718 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
720 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
721 "port %"PRIu16, ofproto->name, ofp_port);
725 return (ofproto->ofproto_class->get_stp_port_status
726 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
730 /* Queue DSCP configuration. */
732 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
733 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
734 * to implement QoS. Instead, it is used to implement features which require
735 * knowledge of what queues exist on a port, and some basic information about
738 * Returns 0 if successful, otherwise a positive errno value. */
740 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
741 const struct ofproto_port_queue *queues,
744 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
747 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
748 ofproto->name, ofp_port);
752 return (ofproto->ofproto_class->set_queues
753 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
757 /* Connectivity Fault Management configuration. */
759 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
761 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
763 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
764 if (ofport && ofproto->ofproto_class->set_cfm) {
765 ofproto->ofproto_class->set_cfm(ofport, NULL);
769 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
770 * basic configuration from the configuration members in 'cfm', and the remote
771 * maintenance point ID from remote_mpid. Ignores the statistics members of
774 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
776 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
777 const struct cfm_settings *s)
779 struct ofport *ofport;
782 ofport = ofproto_get_port(ofproto, ofp_port);
784 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
785 ofproto->name, ofp_port);
789 /* XXX: For configuration simplicity, we only support one remote_mpid
790 * outside of the CFM module. It's not clear if this is the correct long
791 * term solution or not. */
792 error = (ofproto->ofproto_class->set_cfm
793 ? ofproto->ofproto_class->set_cfm(ofport, s)
796 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
797 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
802 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
803 * 'ofproto' does not have a port 'ofp_port'. */
805 ofproto_port_set_bfd(struct ofproto *ofproto, uint16_t ofp_port,
806 const struct smap *cfg)
808 struct ofport *ofport;
811 ofport = ofproto_get_port(ofproto, ofp_port);
813 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
814 ofproto->name, ofp_port);
818 error = (ofproto->ofproto_class->set_bfd
819 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
822 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
823 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
828 /* Populates 'status' with key value pairs indicating the status of the BFD
829 * session on 'ofp_port'. This information is intended to be populated in the
830 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
833 ofproto_port_get_bfd_status(struct ofproto *ofproto, uint16_t ofp_port,
836 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
837 return (ofport && ofproto->ofproto_class->get_bfd_status
838 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
842 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
843 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
844 * 0 if LACP partner information is not current (generally indicating a
845 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
847 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
849 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
850 return (ofport && ofproto->ofproto_class->port_is_lacp_current
851 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
857 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
858 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
859 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
860 * configuration plus, if there is more than one slave, a bonding
863 * If 'aux' is already registered then this function updates its configuration
864 * to 's'. Otherwise, this function registers a new bundle.
866 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
869 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
870 const struct ofproto_bundle_settings *s)
872 return (ofproto->ofproto_class->bundle_set
873 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
877 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
878 * If no such bundle has been registered, this has no effect. */
880 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
882 return ofproto_bundle_register(ofproto, aux, NULL);
886 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
887 * If 'aux' is already registered then this function updates its configuration
888 * to 's'. Otherwise, this function registers a new mirror. */
890 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
891 const struct ofproto_mirror_settings *s)
893 return (ofproto->ofproto_class->mirror_set
894 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
898 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
899 * If no mirror has been registered, this has no effect. */
901 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
903 return ofproto_mirror_register(ofproto, aux, NULL);
906 /* Retrieves statistics from mirror associated with client data pointer
907 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
908 * 'bytes', respectively. If a particular counters is not supported,
909 * the appropriate argument is set to UINT64_MAX. */
911 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
912 uint64_t *packets, uint64_t *bytes)
914 if (!ofproto->ofproto_class->mirror_get_stats) {
915 *packets = *bytes = UINT64_MAX;
919 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
923 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
924 * which all packets are flooded, instead of using MAC learning. If
925 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
927 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
930 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
932 return (ofproto->ofproto_class->set_flood_vlans
933 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
937 /* Returns true if 'aux' is a registered bundle that is currently in use as the
938 * output for a mirror. */
940 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
942 return (ofproto->ofproto_class->is_mirror_output_bundle
943 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
947 /* Configuration of OpenFlow tables. */
949 /* Returns the number of OpenFlow tables in 'ofproto'. */
951 ofproto_get_n_tables(const struct ofproto *ofproto)
953 return ofproto->n_tables;
956 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
957 * settings from 's'. 'table_id' must be in the range 0 through the number of
958 * OpenFlow tables in 'ofproto' minus 1, inclusive.
960 * For read-only tables, only the name may be configured. */
962 ofproto_configure_table(struct ofproto *ofproto, int table_id,
963 const struct ofproto_table_settings *s)
965 struct oftable *table;
967 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
968 table = &ofproto->tables[table_id];
970 oftable_set_name(table, s->name);
972 if (table->flags & OFTABLE_READONLY) {
977 oftable_enable_eviction(table, s->groups, s->n_groups);
979 oftable_disable_eviction(table);
982 table->max_flows = s->max_flows;
983 if (classifier_count(&table->cls) > table->max_flows
984 && table->eviction_fields) {
985 /* 'table' contains more flows than allowed. We might not be able to
986 * evict them right away because of the asynchronous nature of flow
987 * table changes. Schedule eviction for later. */
988 switch (ofproto->state) {
990 ofproto->state = S_EVICT;
994 /* We're already deleting flows, nothing more to do. */
1001 ofproto_has_snoops(const struct ofproto *ofproto)
1003 return connmgr_has_snoops(ofproto->connmgr);
1007 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1009 connmgr_get_snoops(ofproto->connmgr, snoops);
1013 ofproto_flush__(struct ofproto *ofproto)
1015 struct ofopgroup *group;
1016 struct oftable *table;
1018 if (ofproto->ofproto_class->flush) {
1019 ofproto->ofproto_class->flush(ofproto);
1022 group = ofopgroup_create_unattached(ofproto);
1023 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1024 struct rule *rule, *next_rule;
1025 struct cls_cursor cursor;
1027 if (table->flags & OFTABLE_HIDDEN) {
1031 cls_cursor_init(&cursor, &table->cls, NULL);
1032 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1033 if (!rule->pending) {
1034 ofoperation_create(group, rule, OFOPERATION_DELETE,
1036 oftable_remove_rule(rule);
1037 ofproto->ofproto_class->rule_destruct(rule);
1041 ofopgroup_submit(group);
1045 ofproto_destroy__(struct ofproto *ofproto)
1047 struct oftable *table;
1049 ovs_assert(list_is_empty(&ofproto->pending));
1050 ovs_assert(!ofproto->n_pending);
1052 connmgr_destroy(ofproto->connmgr);
1054 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1055 free(ofproto->name);
1056 free(ofproto->type);
1057 free(ofproto->mfr_desc);
1058 free(ofproto->hw_desc);
1059 free(ofproto->sw_desc);
1060 free(ofproto->serial_desc);
1061 free(ofproto->dp_desc);
1062 hmap_destroy(&ofproto->ports);
1063 shash_destroy(&ofproto->port_by_name);
1064 bitmap_free(ofproto->ofp_port_ids);
1065 simap_destroy(&ofproto->ofp_requests);
1067 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1068 oftable_destroy(table);
1070 free(ofproto->tables);
1072 hmap_destroy(&ofproto->deletions);
1074 free(ofproto->vlan_bitmap);
1076 ofproto->ofproto_class->dealloc(ofproto);
1080 ofproto_destroy(struct ofproto *p)
1082 struct ofport *ofport, *next_ofport;
1089 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1090 ofport_destroy(ofport);
1093 p->ofproto_class->destruct(p);
1094 ofproto_destroy__(p);
1097 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1098 * kernel datapath, for example, this destroys the datapath in the kernel, and
1099 * with the netdev-based datapath, it tears down the data structures that
1100 * represent the datapath.
1102 * The datapath should not be currently open as an ofproto. */
1104 ofproto_delete(const char *name, const char *type)
1106 const struct ofproto_class *class = ofproto_class_find__(type);
1107 return (!class ? EAFNOSUPPORT
1108 : !class->del ? EACCES
1109 : class->del(type, name));
1113 process_port_change(struct ofproto *ofproto, int error, char *devname)
1115 if (error == ENOBUFS) {
1116 reinit_ports(ofproto);
1117 } else if (!error) {
1118 update_port(ofproto, devname);
1124 ofproto_type_run(const char *datapath_type)
1126 const struct ofproto_class *class;
1129 datapath_type = ofproto_normalize_type(datapath_type);
1130 class = ofproto_class_find__(datapath_type);
1132 error = class->type_run ? class->type_run(datapath_type) : 0;
1133 if (error && error != EAGAIN) {
1134 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1135 datapath_type, strerror(error));
1141 ofproto_type_run_fast(const char *datapath_type)
1143 const struct ofproto_class *class;
1146 datapath_type = ofproto_normalize_type(datapath_type);
1147 class = ofproto_class_find__(datapath_type);
1149 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1150 if (error && error != EAGAIN) {
1151 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1152 datapath_type, strerror(error));
1158 ofproto_type_wait(const char *datapath_type)
1160 const struct ofproto_class *class;
1162 datapath_type = ofproto_normalize_type(datapath_type);
1163 class = ofproto_class_find__(datapath_type);
1165 if (class->type_wait) {
1166 class->type_wait(datapath_type);
1171 ofproto_run(struct ofproto *p)
1173 struct sset changed_netdevs;
1174 const char *changed_netdev;
1175 struct ofport *ofport;
1178 error = p->ofproto_class->run(p);
1179 if (error && error != EAGAIN) {
1180 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1183 if (p->ofproto_class->port_poll) {
1186 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1187 process_port_change(p, error, devname);
1191 /* Update OpenFlow port status for any port whose netdev has changed.
1193 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1194 * destroyed, so it's not safe to update ports directly from the
1195 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1196 * need this two-phase approach. */
1197 sset_init(&changed_netdevs);
1198 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1199 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1200 if (ofport->change_seq != change_seq) {
1201 ofport->change_seq = change_seq;
1202 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1205 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1206 update_port(p, changed_netdev);
1208 sset_destroy(&changed_netdevs);
1212 connmgr_run(p->connmgr, handle_openflow);
1216 connmgr_run(p->connmgr, NULL);
1218 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1219 p->state = S_OPENFLOW;
1224 connmgr_run(p->connmgr, NULL);
1226 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1227 connmgr_flushed(p->connmgr);
1228 p->state = S_OPENFLOW;
1236 if (time_msec() >= p->next_op_report) {
1237 long long int ago = (time_msec() - p->first_op) / 1000;
1238 long long int interval = (p->last_op - p->first_op) / 1000;
1242 ds_put_format(&s, "%d flow_mods ",
1243 p->n_add + p->n_delete + p->n_modify);
1244 if (interval == ago) {
1245 ds_put_format(&s, "in the last %lld s", ago);
1246 } else if (interval) {
1247 ds_put_format(&s, "in the %lld s starting %lld s ago",
1250 ds_put_format(&s, "%lld s ago", ago);
1253 ds_put_cstr(&s, " (");
1255 ds_put_format(&s, "%d adds, ", p->n_add);
1258 ds_put_format(&s, "%d deletes, ", p->n_delete);
1261 ds_put_format(&s, "%d modifications, ", p->n_modify);
1264 ds_put_char(&s, ')');
1266 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1269 p->n_add = p->n_delete = p->n_modify = 0;
1270 p->next_op_report = LLONG_MAX;
1276 /* Performs periodic activity required by 'ofproto' that needs to be done
1277 * with the least possible latency.
1279 * It makes sense to call this function a couple of times per poll loop, to
1280 * provide a significant performance boost on some benchmarks with the
1281 * ofproto-dpif implementation. */
1283 ofproto_run_fast(struct ofproto *p)
1287 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1288 if (error && error != EAGAIN) {
1289 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1290 p->name, strerror(error));
1296 ofproto_wait(struct ofproto *p)
1298 struct ofport *ofport;
1300 p->ofproto_class->wait(p);
1301 if (p->ofproto_class->port_poll_wait) {
1302 p->ofproto_class->port_poll_wait(p);
1305 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1306 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1307 poll_immediate_wake();
1313 connmgr_wait(p->connmgr, true);
1318 connmgr_wait(p->connmgr, false);
1319 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1320 poll_immediate_wake();
1327 ofproto_is_alive(const struct ofproto *p)
1329 return connmgr_has_controllers(p->connmgr);
1332 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1333 * memory_report(). */
1335 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1337 const struct oftable *table;
1338 unsigned int n_rules;
1340 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1341 simap_increase(usage, "ops",
1342 ofproto->n_pending + hmap_count(&ofproto->deletions));
1345 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1346 n_rules += classifier_count(&table->cls);
1348 simap_increase(usage, "rules", n_rules);
1350 if (ofproto->ofproto_class->get_memory_usage) {
1351 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1354 connmgr_get_memory_usage(ofproto->connmgr, usage);
1358 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1361 connmgr_get_controller_info(ofproto->connmgr, info);
1365 ofproto_free_ofproto_controller_info(struct shash *info)
1367 connmgr_free_controller_info(info);
1370 /* Makes a deep copy of 'old' into 'port'. */
1372 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1374 port->name = xstrdup(old->name);
1375 port->type = xstrdup(old->type);
1376 port->ofp_port = old->ofp_port;
1379 /* Frees memory allocated to members of 'ofproto_port'.
1381 * Do not call this function on an ofproto_port obtained from
1382 * ofproto_port_dump_next(): that function retains ownership of the data in the
1385 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1387 free(ofproto_port->name);
1388 free(ofproto_port->type);
1391 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1393 * This function provides no status indication. An error status for the entire
1394 * dump operation is provided when it is completed by calling
1395 * ofproto_port_dump_done().
1398 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1399 const struct ofproto *ofproto)
1401 dump->ofproto = ofproto;
1402 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1406 /* Attempts to retrieve another port from 'dump', which must have been created
1407 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1408 * 'port' and returns true. On failure, returns false.
1410 * Failure might indicate an actual error or merely that the last port has been
1411 * dumped. An error status for the entire dump operation is provided when it
1412 * is completed by calling ofproto_port_dump_done().
1414 * The ofproto owns the data stored in 'port'. It will remain valid until at
1415 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1416 * ofproto_port_dump_done(). */
1418 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1419 struct ofproto_port *port)
1421 const struct ofproto *ofproto = dump->ofproto;
1427 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1430 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1436 /* Completes port table dump operation 'dump', which must have been created
1437 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1438 * error-free, otherwise a positive errno value describing the problem. */
1440 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1442 const struct ofproto *ofproto = dump->ofproto;
1444 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1447 return dump->error == EOF ? 0 : dump->error;
1450 /* Returns the type to pass to netdev_open() when a datapath of type
1451 * 'datapath_type' has a port of type 'port_type', for a few special
1452 * cases when a netdev type differs from a port type. For example, when
1453 * using the userspace datapath, a port of type "internal" needs to be
1456 * Returns either 'type' itself or a string literal, which must not be
1459 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1461 const struct ofproto_class *class;
1463 datapath_type = ofproto_normalize_type(datapath_type);
1464 class = ofproto_class_find__(datapath_type);
1469 return (class->port_open_type
1470 ? class->port_open_type(datapath_type, port_type)
1474 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1475 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1476 * the port's OpenFlow port number.
1478 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1479 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1480 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1481 * 'ofp_portp' is non-null). */
1483 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1484 uint16_t *ofp_portp)
1486 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1489 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1491 const char *netdev_name = netdev_get_name(netdev);
1493 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1494 update_port(ofproto, netdev_name);
1497 struct ofproto_port ofproto_port;
1499 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1501 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1502 ofproto_port_destroy(&ofproto_port);
1507 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1508 * initializes '*port' appropriately; on failure, returns a positive errno
1511 * The caller owns the data in 'ofproto_port' and must free it with
1512 * ofproto_port_destroy() when it is no longer needed. */
1514 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1515 struct ofproto_port *port)
1519 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1521 memset(port, 0, sizeof *port);
1526 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1527 * Returns 0 if successful, otherwise a positive errno. */
1529 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1531 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1532 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1533 struct simap_node *ofp_request_node;
1536 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1537 if (ofp_request_node) {
1538 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1541 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1542 if (!error && ofport) {
1543 /* 'name' is the netdev's name and update_port() is going to close the
1544 * netdev. Just in case update_port() refers to 'name' after it
1545 * destroys 'ofport', make a copy of it around the update_port()
1547 char *devname = xstrdup(name);
1548 update_port(ofproto, devname);
1554 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1555 * performs the 'n_actions' actions in 'actions'. The new flow will not
1558 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1559 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1560 * controllers; otherwise, it will be hidden.
1562 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1564 * This is a helper function for in-band control and fail-open. */
1566 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1567 unsigned int priority,
1568 const struct ofpact *ofpacts, size_t ofpacts_len)
1570 const struct rule *rule;
1572 rule = rule_from_cls_rule(classifier_find_match_exactly(
1573 &ofproto->tables[0].cls, match, priority));
1574 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1575 ofpacts, ofpacts_len)) {
1576 struct ofputil_flow_mod fm;
1578 memset(&fm, 0, sizeof fm);
1580 fm.priority = priority;
1581 fm.buffer_id = UINT32_MAX;
1582 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1583 fm.ofpacts_len = ofpacts_len;
1584 add_flow(ofproto, NULL, &fm, NULL);
1589 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1590 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1591 * operation cannot be initiated now but may be retried later.
1593 * This is a helper function for in-band control and fail-open. */
1595 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1597 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1600 /* Searches for a rule with matching criteria exactly equal to 'target' in
1601 * ofproto's table 0 and, if it finds one, deletes it.
1603 * This is a helper function for in-band control and fail-open. */
1605 ofproto_delete_flow(struct ofproto *ofproto,
1606 const struct match *target, unsigned int priority)
1610 rule = rule_from_cls_rule(classifier_find_match_exactly(
1611 &ofproto->tables[0].cls, target, priority));
1613 /* No such rule -> success. */
1615 } else if (rule->pending) {
1616 /* An operation on the rule is already pending -> failure.
1617 * Caller must retry later if it's important. */
1620 /* Initiate deletion -> success. */
1621 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1622 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1623 oftable_remove_rule(rule);
1624 ofproto->ofproto_class->rule_destruct(rule);
1625 ofopgroup_submit(group);
1631 /* Starts the process of deleting all of the flows from all of ofproto's flow
1632 * tables and then reintroducing the flows required by in-band control and
1633 * fail-open. The process will complete in a later call to ofproto_run(). */
1635 ofproto_flush_flows(struct ofproto *ofproto)
1637 COVERAGE_INC(ofproto_flush);
1638 ofproto->state = S_FLUSH;
1642 reinit_ports(struct ofproto *p)
1644 struct ofproto_port_dump dump;
1645 struct sset devnames;
1646 struct ofport *ofport;
1647 struct ofproto_port ofproto_port;
1648 const char *devname;
1650 COVERAGE_INC(ofproto_reinit_ports);
1652 sset_init(&devnames);
1653 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1654 sset_add(&devnames, netdev_get_name(ofport->netdev));
1656 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1657 sset_add(&devnames, ofproto_port.name);
1660 SSET_FOR_EACH (devname, &devnames) {
1661 update_port(p, devname);
1663 sset_destroy(&devnames);
1667 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1670 uint16_t end_port_no = ofproto->alloc_port_no;
1672 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1673 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1675 if (ofp_port >= ofproto->max_ports
1676 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1677 /* Search for a free OpenFlow port number. We try not to
1678 * immediately reuse them to prevent problems due to old
1681 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1682 ofproto->alloc_port_no = 0;
1684 if (!bitmap_is_set(ofproto->ofp_port_ids,
1685 ofproto->alloc_port_no)) {
1686 ofp_port = ofproto->alloc_port_no;
1689 if (ofproto->alloc_port_no == end_port_no) {
1694 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1699 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1701 if (ofp_port < ofproto->max_ports) {
1702 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1706 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1707 * pointer if the netdev cannot be opened. On success, also fills in
1709 static struct netdev *
1710 ofport_open(struct ofproto *ofproto,
1711 struct ofproto_port *ofproto_port,
1712 struct ofputil_phy_port *pp)
1714 enum netdev_flags flags;
1715 struct netdev *netdev;
1718 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1720 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1721 "cannot be opened (%s)",
1723 ofproto_port->name, ofproto_port->ofp_port,
1724 ofproto_port->name, strerror(error));
1728 if (ofproto_port->ofp_port == OFPP_NONE) {
1729 if (!strcmp(ofproto->name, ofproto_port->name)) {
1730 ofproto_port->ofp_port = OFPP_LOCAL;
1732 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1733 ofproto_port->name);
1736 pp->port_no = ofproto_port->ofp_port;
1737 netdev_get_etheraddr(netdev, pp->hw_addr);
1738 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1739 netdev_get_flags(netdev, &flags);
1740 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1741 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1742 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1743 &pp->supported, &pp->peer);
1744 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1745 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1750 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1751 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1754 ofport_equal(const struct ofputil_phy_port *a,
1755 const struct ofputil_phy_port *b)
1757 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1758 && a->state == b->state
1759 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1760 && a->curr == b->curr
1761 && a->advertised == b->advertised
1762 && a->supported == b->supported
1763 && a->peer == b->peer
1764 && a->curr_speed == b->curr_speed
1765 && a->max_speed == b->max_speed);
1768 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1769 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1770 * one with the same name or port number). */
1772 ofport_install(struct ofproto *p,
1773 struct netdev *netdev, const struct ofputil_phy_port *pp)
1775 const char *netdev_name = netdev_get_name(netdev);
1776 struct ofport *ofport;
1779 /* Create ofport. */
1780 ofport = p->ofproto_class->port_alloc();
1785 ofport->ofproto = p;
1786 ofport->netdev = netdev;
1787 ofport->change_seq = netdev_change_seq(netdev);
1789 ofport->ofp_port = pp->port_no;
1791 /* Add port to 'p'. */
1792 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1793 shash_add(&p->port_by_name, netdev_name, ofport);
1795 update_mtu(p, ofport);
1797 /* Let the ofproto_class initialize its private data. */
1798 error = p->ofproto_class->port_construct(ofport);
1802 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1806 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1807 p->name, netdev_name, strerror(error));
1809 ofport_destroy__(ofport);
1811 netdev_close(netdev);
1815 /* Removes 'ofport' from 'p' and destroys it. */
1817 ofport_remove(struct ofport *ofport)
1819 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1821 ofport_destroy(ofport);
1824 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1827 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1829 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1831 ofport_remove(port);
1835 /* Updates 'port' with new 'pp' description.
1837 * Does not handle a name or port number change. The caller must implement
1838 * such a change as a delete followed by an add. */
1840 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1842 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1843 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1844 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1845 port->pp.state = pp->state;
1846 port->pp.curr = pp->curr;
1847 port->pp.advertised = pp->advertised;
1848 port->pp.supported = pp->supported;
1849 port->pp.peer = pp->peer;
1850 port->pp.curr_speed = pp->curr_speed;
1851 port->pp.max_speed = pp->max_speed;
1853 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1856 /* Update OpenFlow 'state' in 'port' and notify controller. */
1858 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1860 if (port->pp.state != state) {
1861 port->pp.state = state;
1862 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1868 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1870 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1872 if (port->ofproto->ofproto_class->set_realdev) {
1873 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1875 if (port->ofproto->ofproto_class->set_stp_port) {
1876 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1878 if (port->ofproto->ofproto_class->set_cfm) {
1879 port->ofproto->ofproto_class->set_cfm(port, NULL);
1881 if (port->ofproto->ofproto_class->bundle_remove) {
1882 port->ofproto->ofproto_class->bundle_remove(port);
1888 ofport_destroy__(struct ofport *port)
1890 struct ofproto *ofproto = port->ofproto;
1891 const char *name = netdev_get_name(port->netdev);
1893 hmap_remove(&ofproto->ports, &port->hmap_node);
1894 shash_delete(&ofproto->port_by_name,
1895 shash_find(&ofproto->port_by_name, name));
1897 netdev_close(port->netdev);
1898 ofproto->ofproto_class->port_dealloc(port);
1902 ofport_destroy(struct ofport *port)
1905 dealloc_ofp_port(port->ofproto, port->ofp_port);
1906 port->ofproto->ofproto_class->port_destruct(port);
1907 ofport_destroy__(port);
1912 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1914 struct ofport *port;
1916 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1917 hash_int(ofp_port, 0), &ofproto->ports) {
1918 if (port->ofp_port == ofp_port) {
1926 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1928 struct ofproto *ofproto = port->ofproto;
1931 if (ofproto->ofproto_class->port_get_stats) {
1932 error = ofproto->ofproto_class->port_get_stats(port, stats);
1941 update_port(struct ofproto *ofproto, const char *name)
1943 struct ofproto_port ofproto_port;
1944 struct ofputil_phy_port pp;
1945 struct netdev *netdev;
1946 struct ofport *port;
1948 COVERAGE_INC(ofproto_update_port);
1950 /* Fetch 'name''s location and properties from the datapath. */
1951 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1952 ? ofport_open(ofproto, &ofproto_port, &pp)
1955 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1956 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1957 struct netdev *old_netdev = port->netdev;
1959 /* 'name' hasn't changed location. Any properties changed? */
1960 if (!ofport_equal(&port->pp, &pp)) {
1961 ofport_modified(port, &pp);
1964 update_mtu(ofproto, port);
1966 /* Install the newly opened netdev in case it has changed.
1967 * Don't close the old netdev yet in case port_modified has to
1968 * remove a retained reference to it.*/
1969 port->netdev = netdev;
1970 port->change_seq = netdev_change_seq(netdev);
1972 if (port->ofproto->ofproto_class->port_modified) {
1973 port->ofproto->ofproto_class->port_modified(port);
1976 netdev_close(old_netdev);
1978 /* If 'port' is nonnull then its name differs from 'name' and thus
1979 * we should delete it. If we think there's a port named 'name'
1980 * then its port number must be wrong now so delete it too. */
1982 ofport_remove(port);
1984 ofport_remove_with_name(ofproto, name);
1985 ofport_install(ofproto, netdev, &pp);
1988 /* Any port named 'name' is gone now. */
1989 ofport_remove_with_name(ofproto, name);
1991 ofproto_port_destroy(&ofproto_port);
1995 init_ports(struct ofproto *p)
1997 struct ofproto_port_dump dump;
1998 struct ofproto_port ofproto_port;
1999 struct shash_node *node, *next;
2001 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2002 const char *name = ofproto_port.name;
2004 if (shash_find(&p->port_by_name, name)) {
2005 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2008 struct ofputil_phy_port pp;
2009 struct netdev *netdev;
2011 /* Check if an OpenFlow port number had been requested. */
2012 node = shash_find(&init_ofp_ports, name);
2014 const struct iface_hint *iface_hint = node->data;
2015 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
2018 netdev = ofport_open(p, &ofproto_port, &pp);
2020 ofport_install(p, netdev, &pp);
2025 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2026 struct iface_hint *iface_hint = node->data;
2028 if (!strcmp(iface_hint->br_name, p->name)) {
2029 free(iface_hint->br_name);
2030 free(iface_hint->br_type);
2032 shash_delete(&init_ofp_ports, node);
2039 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2040 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2042 find_min_mtu(struct ofproto *p)
2044 struct ofport *ofport;
2047 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2048 struct netdev *netdev = ofport->netdev;
2051 /* Skip any internal ports, since that's what we're trying to
2053 if (!strcmp(netdev_get_type(netdev), "internal")) {
2057 if (netdev_get_mtu(netdev, &dev_mtu)) {
2060 if (!mtu || dev_mtu < mtu) {
2065 return mtu ? mtu: ETH_PAYLOAD_MAX;
2068 /* Update MTU of all datapath devices on 'p' to the minimum of the
2069 * non-datapath ports in event of 'port' added or changed. */
2071 update_mtu(struct ofproto *p, struct ofport *port)
2073 struct ofport *ofport;
2074 struct netdev *netdev = port->netdev;
2075 int dev_mtu, old_min;
2077 if (netdev_get_mtu(netdev, &dev_mtu)) {
2081 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2082 if (dev_mtu > p->min_mtu) {
2083 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2084 dev_mtu = p->min_mtu;
2087 port->mtu = dev_mtu;
2091 /* For non-internal port find new min mtu. */
2092 old_min = p->min_mtu;
2093 port->mtu = dev_mtu;
2094 p->min_mtu = find_min_mtu(p);
2095 if (p->min_mtu == old_min) {
2099 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2100 struct netdev *netdev = ofport->netdev;
2102 if (!strcmp(netdev_get_type(netdev), "internal")) {
2103 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2104 ofport->mtu = p->min_mtu;
2111 ofproto_rule_destroy__(struct rule *rule)
2114 cls_rule_destroy(&rule->cr);
2115 free(rule->ofpacts);
2116 rule->ofproto->ofproto_class->rule_dealloc(rule);
2120 /* This function allows an ofproto implementation to destroy any rules that
2121 * remain when its ->destruct() function is called. The caller must have
2122 * already uninitialized any derived members of 'rule' (step 5 described in the
2123 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2124 * This function implements steps 6 and 7.
2126 * This function should only be called from an ofproto implementation's
2127 * ->destruct() function. It is not suitable elsewhere. */
2129 ofproto_rule_destroy(struct rule *rule)
2131 ovs_assert(!rule->pending);
2132 oftable_remove_rule(rule);
2133 ofproto_rule_destroy__(rule);
2136 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2137 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2139 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2141 return (port == OFPP_ANY
2142 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2145 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2146 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2148 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2150 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2155 case OFOPERATION_ADD:
2156 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2158 case OFOPERATION_DELETE:
2161 case OFOPERATION_MODIFY:
2162 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2168 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2169 * statistics appropriately. 'packet' must have at least sizeof(struct
2170 * ofp10_packet_in) bytes of headroom.
2172 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2173 * with statistics for 'packet' either way.
2175 * Takes ownership of 'packet'. */
2177 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2181 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2183 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2184 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2187 /* Returns true if 'rule' should be hidden from the controller.
2189 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2190 * (e.g. by in-band control) and are intentionally hidden from the
2193 ofproto_rule_is_hidden(const struct rule *rule)
2195 return rule->cr.priority > UINT16_MAX;
2198 static enum oftable_flags
2199 rule_get_flags(const struct rule *rule)
2201 return rule->ofproto->tables[rule->table_id].flags;
2205 rule_is_modifiable(const struct rule *rule)
2207 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2211 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2213 ofconn_send_reply(ofconn, make_echo_reply(oh));
2218 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2220 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2221 struct ofputil_switch_features features;
2222 struct ofport *port;
2228 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2230 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2232 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2233 * if present, are always at the end.) */
2234 n_tables = ofproto->n_tables;
2235 for (i = 0; i < ofproto->n_tables; i++) {
2236 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2242 features.datapath_id = ofproto->datapath_id;
2243 features.n_buffers = pktbuf_capacity();
2244 features.n_tables = n_tables;
2245 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2246 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2248 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2250 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2251 features.auxiliary_id = 0;
2252 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2254 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2255 ofputil_put_switch_features_port(&port->pp, b);
2258 ofconn_send_reply(ofconn, b);
2263 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2265 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2266 struct ofp_switch_config *osc;
2267 enum ofp_config_flags flags;
2271 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2272 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2273 flags = ofproto->frag_handling;
2274 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2275 if (oh->version < OFP13_VERSION
2276 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2277 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2279 osc->flags = htons(flags);
2280 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2281 ofconn_send_reply(ofconn, buf);
2287 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2289 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2290 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2291 uint16_t flags = ntohs(osc->flags);
2293 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2294 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2295 enum ofp_config_flags cur = ofproto->frag_handling;
2296 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2298 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2300 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2301 ofproto->frag_handling = next;
2303 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2305 ofputil_frag_handling_to_string(next));
2309 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2310 ofconn_set_invalid_ttl_to_controller(ofconn,
2311 (oh->version < OFP13_VERSION
2312 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2314 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2319 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2320 * error message code for the caller to propagate upward. Otherwise, returns
2323 * The log message mentions 'msg_type'. */
2325 reject_slave_controller(struct ofconn *ofconn)
2327 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2328 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2329 return OFPERR_OFPBRC_EPERM;
2336 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2338 struct ofproto *p = ofconn_get_ofproto(ofconn);
2339 struct ofputil_packet_out po;
2340 struct ofpbuf *payload;
2341 uint64_t ofpacts_stub[1024 / 8];
2342 struct ofpbuf ofpacts;
2346 COVERAGE_INC(ofproto_packet_out);
2348 error = reject_slave_controller(ofconn);
2353 /* Decode message. */
2354 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2355 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2357 goto exit_free_ofpacts;
2359 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2360 error = OFPERR_OFPBRC_BAD_PORT;
2361 goto exit_free_ofpacts;
2366 if (po.buffer_id != UINT32_MAX) {
2367 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2368 if (error || !payload) {
2369 goto exit_free_ofpacts;
2372 payload = xmalloc(sizeof *payload);
2373 ofpbuf_use_const(payload, po.packet, po.packet_len);
2376 /* Verify actions against packet, then send packet if successful. */
2377 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2378 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2380 error = p->ofproto_class->packet_out(p, payload, &flow,
2381 po.ofpacts, po.ofpacts_len);
2383 ofpbuf_delete(payload);
2386 ofpbuf_uninit(&ofpacts);
2392 update_port_config(struct ofport *port,
2393 enum ofputil_port_config config,
2394 enum ofputil_port_config mask)
2396 enum ofputil_port_config old_config = port->pp.config;
2397 enum ofputil_port_config toggle;
2399 toggle = (config ^ port->pp.config) & mask;
2400 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2401 if (config & OFPUTIL_PC_PORT_DOWN) {
2402 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2404 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2406 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2409 port->pp.config ^= toggle;
2410 if (port->pp.config != old_config) {
2411 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2416 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2418 struct ofproto *p = ofconn_get_ofproto(ofconn);
2419 struct ofputil_port_mod pm;
2420 struct ofport *port;
2423 error = reject_slave_controller(ofconn);
2428 error = ofputil_decode_port_mod(oh, &pm);
2433 port = ofproto_get_port(p, pm.port_no);
2435 return OFPERR_OFPPMFC_BAD_PORT;
2436 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2437 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2439 update_port_config(port, pm.config, pm.mask);
2441 netdev_set_advertisements(port->netdev, pm.advertise);
2448 handle_desc_stats_request(struct ofconn *ofconn,
2449 const struct ofp_header *request)
2451 static const char *default_mfr_desc = "Nicira, Inc.";
2452 static const char *default_hw_desc = "Open vSwitch";
2453 static const char *default_sw_desc = VERSION;
2454 static const char *default_serial_desc = "None";
2455 static const char *default_dp_desc = "None";
2457 struct ofproto *p = ofconn_get_ofproto(ofconn);
2458 struct ofp_desc_stats *ods;
2461 msg = ofpraw_alloc_stats_reply(request, 0);
2462 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2463 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2464 sizeof ods->mfr_desc);
2465 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2466 sizeof ods->hw_desc);
2467 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2468 sizeof ods->sw_desc);
2469 ovs_strlcpy(ods->serial_num,
2470 p->serial_desc ? p->serial_desc : default_serial_desc,
2471 sizeof ods->serial_num);
2472 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2473 sizeof ods->dp_desc);
2474 ofconn_send_reply(ofconn, msg);
2480 handle_table_stats_request(struct ofconn *ofconn,
2481 const struct ofp_header *request)
2483 struct ofproto *p = ofconn_get_ofproto(ofconn);
2484 struct ofp12_table_stats *ots;
2489 /* Set up default values.
2491 * ofp12_table_stats is used as a generic structure as
2492 * it is able to hold all the fields for ofp10_table_stats
2493 * and ofp11_table_stats (and of course itself).
2495 ots = xcalloc(p->n_tables, sizeof *ots);
2496 for (i = 0; i < p->n_tables; i++) {
2497 ots[i].table_id = i;
2498 sprintf(ots[i].name, "table%zu", i);
2499 ots[i].match = htonll(OFPXMT12_MASK);
2500 ots[i].wildcards = htonll(OFPXMT12_MASK);
2501 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2502 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2503 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2504 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2505 ots[i].metadata_match = htonll(UINT64_MAX);
2506 ots[i].metadata_write = htonll(UINT64_MAX);
2507 ots[i].instructions = htonl(OFPIT11_ALL);
2508 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2509 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2510 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2513 p->ofproto_class->get_tables(p, ots);
2515 /* Post-process the tables, dropping hidden tables. */
2516 n_tables = p->n_tables;
2517 for (i = 0; i < p->n_tables; i++) {
2518 const struct oftable *table = &p->tables[i];
2520 if (table->flags & OFTABLE_HIDDEN) {
2526 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2529 if (table->max_flows < ntohl(ots[i].max_entries)) {
2530 ots[i].max_entries = htonl(table->max_flows);
2534 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2535 ofconn_send_reply(ofconn, msg);
2543 append_port_stat(struct ofport *port, struct list *replies)
2545 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2547 /* Intentionally ignore return value, since errors will set
2548 * 'stats' to all-1s, which is correct for OpenFlow, and
2549 * netdev_get_stats() will log errors. */
2550 ofproto_port_get_stats(port, &ops.stats);
2552 ofputil_append_port_stat(replies, &ops);
2556 handle_port_stats_request(struct ofconn *ofconn,
2557 const struct ofp_header *request)
2559 struct ofproto *p = ofconn_get_ofproto(ofconn);
2560 struct ofport *port;
2561 struct list replies;
2565 error = ofputil_decode_port_stats_request(request, &port_no);
2570 ofpmp_init(&replies, request);
2571 if (port_no != OFPP_ANY) {
2572 port = ofproto_get_port(p, port_no);
2574 append_port_stat(port, &replies);
2577 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2578 append_port_stat(port, &replies);
2582 ofconn_send_replies(ofconn, &replies);
2587 handle_port_desc_stats_request(struct ofconn *ofconn,
2588 const struct ofp_header *request)
2590 struct ofproto *p = ofconn_get_ofproto(ofconn);
2591 enum ofp_version version;
2592 struct ofport *port;
2593 struct list replies;
2595 ofpmp_init(&replies, request);
2597 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2598 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2599 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2602 ofconn_send_replies(ofconn, &replies);
2607 calc_flow_duration__(long long int start, long long int now,
2608 uint32_t *sec, uint32_t *nsec)
2610 long long int msecs = now - start;
2611 *sec = msecs / 1000;
2612 *nsec = (msecs % 1000) * (1000 * 1000);
2615 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2616 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2618 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2620 return (table_id == 0xff || table_id < ofproto->n_tables
2622 : OFPERR_OFPBRC_BAD_TABLE_ID);
2626 static struct oftable *
2627 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2629 struct oftable *table;
2631 for (table = &ofproto->tables[table_id];
2632 table < &ofproto->tables[ofproto->n_tables];
2634 if (!(table->flags & OFTABLE_HIDDEN)) {
2642 static struct oftable *
2643 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2645 if (table_id == 0xff) {
2646 return next_visible_table(ofproto, 0);
2647 } else if (table_id < ofproto->n_tables) {
2648 return &ofproto->tables[table_id];
2654 static struct oftable *
2655 next_matching_table(const struct ofproto *ofproto,
2656 const struct oftable *table, uint8_t table_id)
2658 return (table_id == 0xff
2659 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2663 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2665 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2666 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2668 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2669 * only once, for that table. (This can be used to access tables marked
2672 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2673 * entered at all. (Perhaps you should have validated TABLE_ID with
2674 * check_table_id().)
2676 * All parameters are evaluated multiple times.
2678 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2679 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2681 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2683 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2684 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2685 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2688 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2689 * to 'out_port' are included.
2691 * Hidden rules are always omitted.
2693 * Returns 0 on success, otherwise an OpenFlow error code. */
2695 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2696 const struct match *match,
2697 ovs_be64 cookie, ovs_be64 cookie_mask,
2698 uint16_t out_port, struct list *rules)
2700 struct oftable *table;
2704 error = check_table_id(ofproto, table_id);
2710 cls_rule_init(&cr, match, 0);
2711 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2712 struct cls_cursor cursor;
2715 cls_cursor_init(&cursor, &table->cls, &cr);
2716 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2717 if (rule->pending) {
2718 error = OFPROTO_POSTPONE;
2721 if (!ofproto_rule_is_hidden(rule)
2722 && ofproto_rule_has_out_port(rule, out_port)
2723 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2724 list_push_back(rules, &rule->ofproto_node);
2730 cls_rule_destroy(&cr);
2734 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2735 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2736 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2739 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2740 * to 'out_port' are included.
2742 * Hidden rules are always omitted.
2744 * Returns 0 on success, otherwise an OpenFlow error code. */
2746 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2747 const struct match *match, unsigned int priority,
2748 ovs_be64 cookie, ovs_be64 cookie_mask,
2749 uint16_t out_port, struct list *rules)
2751 struct oftable *table;
2755 error = check_table_id(ofproto, table_id);
2761 cls_rule_init(&cr, match, priority);
2762 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2765 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2768 if (rule->pending) {
2769 error = OFPROTO_POSTPONE;
2772 if (!ofproto_rule_is_hidden(rule)
2773 && ofproto_rule_has_out_port(rule, out_port)
2774 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2775 list_push_back(rules, &rule->ofproto_node);
2781 cls_rule_destroy(&cr);
2785 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2786 * forced into the range of a uint16_t. */
2788 age_secs(long long int age_ms)
2790 return (age_ms < 0 ? 0
2791 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2792 : (unsigned int) age_ms / 1000);
2796 handle_flow_stats_request(struct ofconn *ofconn,
2797 const struct ofp_header *request)
2799 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2800 struct ofputil_flow_stats_request fsr;
2801 struct list replies;
2806 error = ofputil_decode_flow_stats_request(&fsr, request);
2811 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2812 fsr.cookie, fsr.cookie_mask,
2813 fsr.out_port, &rules);
2818 ofpmp_init(&replies, request);
2819 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2820 long long int now = time_msec();
2821 struct ofputil_flow_stats fs;
2823 minimatch_expand(&rule->cr.match, &fs.match);
2824 fs.priority = rule->cr.priority;
2825 fs.cookie = rule->flow_cookie;
2826 fs.table_id = rule->table_id;
2827 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2829 fs.idle_timeout = rule->idle_timeout;
2830 fs.hard_timeout = rule->hard_timeout;
2831 fs.idle_age = age_secs(now - rule->used);
2832 fs.hard_age = age_secs(now - rule->modified);
2833 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2835 fs.ofpacts = rule->ofpacts;
2836 fs.ofpacts_len = rule->ofpacts_len;
2838 if (rule->send_flow_removed) {
2839 fs.flags |= OFPFF_SEND_FLOW_REM;
2840 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2841 and OFPFF13_NO_BYT_COUNTS */
2843 ofputil_append_flow_stats_reply(&fs, &replies);
2845 ofconn_send_replies(ofconn, &replies);
2851 flow_stats_ds(struct rule *rule, struct ds *results)
2853 uint64_t packet_count, byte_count;
2855 rule->ofproto->ofproto_class->rule_get_stats(rule,
2856 &packet_count, &byte_count);
2858 if (rule->table_id != 0) {
2859 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2861 ds_put_format(results, "duration=%llds, ",
2862 (time_msec() - rule->created) / 1000);
2863 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2864 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2865 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2866 cls_rule_format(&rule->cr, results);
2867 ds_put_char(results, ',');
2868 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2869 ds_put_cstr(results, "\n");
2872 /* Adds a pretty-printed description of all flows to 'results', including
2873 * hidden flows (e.g., set up by in-band control). */
2875 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2877 struct oftable *table;
2879 OFPROTO_FOR_EACH_TABLE (table, p) {
2880 struct cls_cursor cursor;
2883 cls_cursor_init(&cursor, &table->cls, NULL);
2884 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2885 flow_stats_ds(rule, results);
2890 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2891 * '*engine_type' and '*engine_id', respectively. */
2893 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2894 uint8_t *engine_type, uint8_t *engine_id)
2896 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2899 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2900 * true if the port's CFM status was successfully stored into '*status'.
2901 * Returns false if the port did not have CFM configured, in which case
2902 * '*status' is indeterminate.
2904 * The caller must provide and owns '*status', but it does not own and must not
2905 * modify or free the array returned in 'status->rmps'. */
2907 ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
2908 struct ofproto_cfm_status *status)
2910 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2912 && ofproto->ofproto_class->get_cfm_status
2913 && ofproto->ofproto_class->get_cfm_status(ofport, status));
2917 handle_aggregate_stats_request(struct ofconn *ofconn,
2918 const struct ofp_header *oh)
2920 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2921 struct ofputil_flow_stats_request request;
2922 struct ofputil_aggregate_stats stats;
2923 bool unknown_packets, unknown_bytes;
2924 struct ofpbuf *reply;
2929 error = ofputil_decode_flow_stats_request(&request, oh);
2934 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2935 request.cookie, request.cookie_mask,
2936 request.out_port, &rules);
2941 memset(&stats, 0, sizeof stats);
2942 unknown_packets = unknown_bytes = false;
2943 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2944 uint64_t packet_count;
2945 uint64_t byte_count;
2947 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2950 if (packet_count == UINT64_MAX) {
2951 unknown_packets = true;
2953 stats.packet_count += packet_count;
2956 if (byte_count == UINT64_MAX) {
2957 unknown_bytes = true;
2959 stats.byte_count += byte_count;
2964 if (unknown_packets) {
2965 stats.packet_count = UINT64_MAX;
2967 if (unknown_bytes) {
2968 stats.byte_count = UINT64_MAX;
2971 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2972 ofconn_send_reply(ofconn, reply);
2977 struct queue_stats_cbdata {
2978 struct ofport *ofport;
2979 struct list replies;
2983 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2984 const struct netdev_queue_stats *stats)
2987 struct ofputil_queue_stats oqs = {
2988 .port_no = cbdata->ofport->pp.port_no,
2989 .queue_id = queue_id,
2992 ofputil_append_queue_stat(&cbdata->replies, &oqs);
2996 handle_queue_stats_dump_cb(uint32_t queue_id,
2997 struct netdev_queue_stats *stats,
3000 struct queue_stats_cbdata *cbdata = cbdata_;
3002 put_queue_stats(cbdata, queue_id, stats);
3006 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3007 struct queue_stats_cbdata *cbdata)
3009 cbdata->ofport = port;
3010 if (queue_id == OFPQ_ALL) {
3011 netdev_dump_queue_stats(port->netdev,
3012 handle_queue_stats_dump_cb, cbdata);
3014 struct netdev_queue_stats stats;
3016 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3017 put_queue_stats(cbdata, queue_id, &stats);
3019 return OFPERR_OFPQOFC_BAD_QUEUE;
3026 handle_queue_stats_request(struct ofconn *ofconn,
3027 const struct ofp_header *rq)
3029 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3030 struct queue_stats_cbdata cbdata;
3031 struct ofport *port;
3033 struct ofputil_queue_stats_request oqsr;
3035 COVERAGE_INC(ofproto_queue_req);
3037 ofpmp_init(&cbdata.replies, rq);
3039 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3044 if (oqsr.port_no == OFPP_ANY) {
3045 error = OFPERR_OFPQOFC_BAD_QUEUE;
3046 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3047 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3052 port = ofproto_get_port(ofproto, oqsr.port_no);
3054 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3055 : OFPERR_OFPQOFC_BAD_PORT);
3058 ofconn_send_replies(ofconn, &cbdata.replies);
3060 ofpbuf_list_delete(&cbdata.replies);
3067 is_flow_deletion_pending(const struct ofproto *ofproto,
3068 const struct cls_rule *cls_rule,
3071 if (!hmap_is_empty(&ofproto->deletions)) {
3072 struct ofoperation *op;
3074 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3075 cls_rule_hash(cls_rule, table_id),
3076 &ofproto->deletions) {
3077 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3086 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3087 * in which no matching flow already exists in the flow table.
3089 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3090 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3091 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3092 * initiated now but may be retried later.
3094 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3095 * ownership remains with the caller.
3097 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3100 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3101 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3103 struct oftable *table;
3104 struct ofopgroup *group;
3105 struct rule *victim;
3110 error = check_table_id(ofproto, fm->table_id);
3116 if (fm->table_id == 0xff) {
3118 if (ofproto->ofproto_class->rule_choose_table) {
3119 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3125 ovs_assert(table_id < ofproto->n_tables);
3126 table = &ofproto->tables[table_id];
3128 table = &ofproto->tables[0];
3130 } else if (fm->table_id < ofproto->n_tables) {
3131 table = &ofproto->tables[fm->table_id];
3133 return OFPERR_OFPBRC_BAD_TABLE_ID;
3136 if (table->flags & OFTABLE_READONLY) {
3137 return OFPERR_OFPBRC_EPERM;
3140 /* Allocate new rule and initialize classifier rule. */
3141 rule = ofproto->ofproto_class->rule_alloc();
3143 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3144 ofproto->name, strerror(error));
3147 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3149 /* Serialize against pending deletion. */
3150 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3151 cls_rule_destroy(&rule->cr);
3152 ofproto->ofproto_class->rule_dealloc(rule);
3153 return OFPROTO_POSTPONE;
3156 /* Check for overlap, if requested. */
3157 if (fm->flags & OFPFF_CHECK_OVERLAP
3158 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3159 cls_rule_destroy(&rule->cr);
3160 ofproto->ofproto_class->rule_dealloc(rule);
3161 return OFPERR_OFPFMFC_OVERLAP;
3164 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3166 rule->ofproto = ofproto;
3167 rule->pending = NULL;
3168 rule->flow_cookie = fm->new_cookie;
3169 rule->created = rule->modified = rule->used = time_msec();
3170 rule->idle_timeout = fm->idle_timeout;
3171 rule->hard_timeout = fm->hard_timeout;
3172 rule->table_id = table - ofproto->tables;
3173 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3174 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3175 and OFPFF13_NO_BYT_COUNTS */
3176 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3177 rule->ofpacts_len = fm->ofpacts_len;
3178 rule->evictable = true;
3179 rule->eviction_group = NULL;
3180 list_init(&rule->expirable);
3181 rule->monitor_flags = 0;
3182 rule->add_seqno = 0;
3183 rule->modify_seqno = 0;
3185 /* Insert new rule. */
3186 victim = oftable_replace_rule(rule);
3187 if (victim && !rule_is_modifiable(victim)) {
3188 error = OFPERR_OFPBRC_EPERM;
3189 } else if (victim && victim->pending) {
3190 error = OFPROTO_POSTPONE;
3192 struct ofoperation *op;
3195 if (classifier_count(&table->cls) > table->max_flows) {
3198 was_evictable = rule->evictable;
3199 rule->evictable = false;
3200 evict = choose_rule_to_evict(table);
3201 rule->evictable = was_evictable;
3204 error = OFPERR_OFPFMFC_TABLE_FULL;
3206 } else if (evict->pending) {
3207 error = OFPROTO_POSTPONE;
3214 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3215 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3216 op->victim = victim;
3218 error = ofproto->ofproto_class->rule_construct(rule);
3220 op->group->n_running--;
3221 ofoperation_destroy(rule->pending);
3223 delete_flow__(evict, group);
3225 ofopgroup_submit(group);
3229 /* Back out if an error occurred. */
3231 oftable_substitute_rule(rule, victim);
3232 ofproto_rule_destroy__(rule);
3237 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3239 /* Modifies the rules listed in 'rules', changing their actions to match those
3242 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3245 * Returns 0 on success, otherwise an OpenFlow error code. */
3247 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3248 const struct ofputil_flow_mod *fm,
3249 const struct ofp_header *request, struct list *rules)
3251 struct ofopgroup *group;
3255 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3256 error = OFPERR_OFPBRC_EPERM;
3257 LIST_FOR_EACH (rule, ofproto_node, rules) {
3258 struct ofoperation *op;
3259 bool actions_changed;
3260 ovs_be64 new_cookie;
3262 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3264 if (rule_is_modifiable(rule)) {
3265 /* At least one rule is modifiable, don't report EPERM error. */
3271 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3272 rule->ofpacts, rule->ofpacts_len);
3273 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3275 : rule->flow_cookie);
3277 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3278 rule->flow_cookie = new_cookie;
3279 if (actions_changed) {
3280 op->ofpacts = rule->ofpacts;
3281 op->ofpacts_len = rule->ofpacts_len;
3282 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3283 rule->ofpacts_len = fm->ofpacts_len;
3284 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3286 ofoperation_complete(op, 0);
3289 ofopgroup_submit(group);
3295 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3296 const struct ofputil_flow_mod *fm,
3297 const struct ofp_header *request)
3299 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3302 return add_flow(ofproto, ofconn, fm, request);
3305 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3308 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3311 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3312 const struct ofputil_flow_mod *fm,
3313 const struct ofp_header *request)
3318 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3319 fm->cookie, fm->cookie_mask,
3323 } else if (list_is_empty(&rules)) {
3324 return modify_flows_add(ofproto, ofconn, fm, request);
3326 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3330 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3333 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3336 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3337 const struct ofputil_flow_mod *fm,
3338 const struct ofp_header *request)
3343 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3344 fm->priority, fm->cookie, fm->cookie_mask,
3349 } else if (list_is_empty(&rules)) {
3350 return modify_flows_add(ofproto, ofconn, fm, request);
3352 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3353 fm, request, &rules)
3358 /* OFPFC_DELETE implementation. */
3361 delete_flow__(struct rule *rule, struct ofopgroup *group)
3363 struct ofproto *ofproto = rule->ofproto;
3365 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3367 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3368 oftable_remove_rule(rule);
3369 ofproto->ofproto_class->rule_destruct(rule);
3372 /* Deletes the rules listed in 'rules'.
3374 * Returns 0 on success, otherwise an OpenFlow error code. */
3376 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3377 const struct ofp_header *request, struct list *rules)
3379 struct rule *rule, *next;
3380 struct ofopgroup *group;
3382 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3383 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3384 delete_flow__(rule, group);
3386 ofopgroup_submit(group);
3391 /* Implements OFPFC_DELETE. */
3393 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3394 const struct ofputil_flow_mod *fm,
3395 const struct ofp_header *request)
3400 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3401 fm->cookie, fm->cookie_mask,
3402 fm->out_port, &rules);
3403 return (error ? error
3404 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3409 /* Implements OFPFC_DELETE_STRICT. */
3411 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3412 const struct ofputil_flow_mod *fm,
3413 const struct ofp_header *request)
3418 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3419 fm->priority, fm->cookie, fm->cookie_mask,
3420 fm->out_port, &rules);
3421 return (error ? error
3422 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3428 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3430 struct ofputil_flow_removed fr;
3432 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3436 minimatch_expand(&rule->cr.match, &fr.match);
3437 fr.priority = rule->cr.priority;
3438 fr.cookie = rule->flow_cookie;
3440 fr.table_id = rule->table_id;
3441 calc_flow_duration__(rule->created, time_msec(),
3442 &fr.duration_sec, &fr.duration_nsec);
3443 fr.idle_timeout = rule->idle_timeout;
3444 fr.hard_timeout = rule->hard_timeout;
3445 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3448 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3452 ofproto_rule_update_used(struct rule *rule, long long int used)
3454 if (used > rule->used) {
3455 struct eviction_group *evg = rule->eviction_group;
3459 heap_change(&evg->rules, &rule->evg_node,
3460 rule_eviction_priority(rule));
3465 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3466 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3469 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3472 * ofproto implementation ->run() functions should use this function to expire
3473 * OpenFlow flows. */
3475 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3477 struct ofproto *ofproto = rule->ofproto;
3478 struct ofopgroup *group;
3480 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3482 ofproto_rule_send_removed(rule, reason);
3484 group = ofopgroup_create_unattached(ofproto);
3485 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3486 oftable_remove_rule(rule);
3487 ofproto->ofproto_class->rule_destruct(rule);
3488 ofopgroup_submit(group);
3492 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3494 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3495 struct ofputil_flow_mod fm;
3496 uint64_t ofpacts_stub[1024 / 8];
3497 struct ofpbuf ofpacts;
3501 error = reject_slave_controller(ofconn);
3506 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3507 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3510 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3511 &fm.match.flow, ofproto->max_ports);
3514 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3517 goto exit_free_ofpacts;
3520 /* Record the operation for logging a summary report. */
3521 switch (fm.command) {
3527 case OFPFC_MODIFY_STRICT:
3528 ofproto->n_modify++;
3532 case OFPFC_DELETE_STRICT:
3533 ofproto->n_delete++;
3538 if (ofproto->next_op_report == LLONG_MAX) {
3539 ofproto->first_op = now;
3540 ofproto->next_op_report = MAX(now + 10 * 1000,
3541 ofproto->op_backoff);
3542 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3544 ofproto->last_op = now;
3547 ofpbuf_uninit(&ofpacts);
3553 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3554 const struct ofputil_flow_mod *fm,
3555 const struct ofp_header *oh)
3557 if (ofproto->n_pending >= 50) {
3558 ovs_assert(!list_is_empty(&ofproto->pending));
3559 return OFPROTO_POSTPONE;
3562 switch (fm->command) {
3564 return add_flow(ofproto, ofconn, fm, oh);
3567 return modify_flows_loose(ofproto, ofconn, fm, oh);
3569 case OFPFC_MODIFY_STRICT:
3570 return modify_flow_strict(ofproto, ofconn, fm, oh);
3573 return delete_flows_loose(ofproto, ofconn, fm, oh);
3575 case OFPFC_DELETE_STRICT:
3576 return delete_flow_strict(ofproto, ofconn, fm, oh);
3579 if (fm->command > 0xff) {
3580 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3581 "flow_mod_table_id extension is not enabled",
3584 return OFPERR_OFPFMFC_BAD_COMMAND;
3589 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3591 struct ofputil_role_request request;
3592 struct ofputil_role_request reply;
3596 error = ofputil_decode_role_message(oh, &request);
3601 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3602 if (ofconn_get_role(ofconn) != request.role
3603 && ofconn_has_pending_opgroups(ofconn)) {
3604 return OFPROTO_POSTPONE;
3607 if (request.have_generation_id
3608 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3609 return OFPERR_OFPRRFC_STALE;
3612 ofconn_set_role(ofconn, request.role);
3615 reply.role = ofconn_get_role(ofconn);
3616 reply.have_generation_id = ofconn_get_master_election_id(
3617 ofconn, &reply.generation_id);
3618 buf = ofputil_encode_role_reply(oh, &reply);
3619 ofconn_send_reply(ofconn, buf);
3625 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3626 const struct ofp_header *oh)
3628 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3629 enum ofputil_protocol cur, next;
3631 cur = ofconn_get_protocol(ofconn);
3632 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3633 ofconn_set_protocol(ofconn, next);
3639 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3641 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3642 enum ofputil_protocol cur, next;
3643 enum ofputil_protocol next_base;
3645 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3647 return OFPERR_OFPBRC_EPERM;
3650 cur = ofconn_get_protocol(ofconn);
3651 next = ofputil_protocol_set_base(cur, next_base);
3652 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3653 /* Avoid sending async messages in surprising protocol. */
3654 return OFPROTO_POSTPONE;
3657 ofconn_set_protocol(ofconn, next);
3662 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3663 const struct ofp_header *oh)
3665 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3668 format = ntohl(msg->format);
3669 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3670 return OFPERR_OFPBRC_EPERM;
3673 if (format != ofconn_get_packet_in_format(ofconn)
3674 && ofconn_has_pending_opgroups(ofconn)) {
3675 /* Avoid sending async message in surprsing packet in format. */
3676 return OFPROTO_POSTPONE;
3679 ofconn_set_packet_in_format(ofconn, format);
3684 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3686 const struct nx_async_config *msg = ofpmsg_body(oh);
3687 uint32_t master[OAM_N_TYPES];
3688 uint32_t slave[OAM_N_TYPES];
3690 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3691 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3692 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3694 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3695 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3696 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3698 ofconn_set_async_config(ofconn, master, slave);
3699 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3700 !ofconn_get_miss_send_len(ofconn)) {
3701 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3708 handle_nxt_set_controller_id(struct ofconn *ofconn,
3709 const struct ofp_header *oh)
3711 const struct nx_controller_id *nci = ofpmsg_body(oh);
3713 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3714 return OFPERR_NXBRC_MUST_BE_ZERO;
3717 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3722 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3726 if (ofconn_has_pending_opgroups(ofconn)) {
3727 return OFPROTO_POSTPONE;
3730 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3731 ? OFPRAW_OFPT10_BARRIER_REPLY
3732 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3733 ofconn_send_reply(ofconn, buf);
3738 ofproto_compose_flow_refresh_update(const struct rule *rule,
3739 enum nx_flow_monitor_flags flags,
3742 struct ofoperation *op = rule->pending;
3743 struct ofputil_flow_update fu;
3746 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3747 /* We'll report the final flow when the operation completes. Reporting
3748 * it now would cause a duplicate report later. */
3752 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3753 ? NXFME_ADDED : NXFME_MODIFIED);
3755 fu.idle_timeout = rule->idle_timeout;
3756 fu.hard_timeout = rule->hard_timeout;
3757 fu.table_id = rule->table_id;
3758 fu.cookie = rule->flow_cookie;
3759 minimatch_expand(&rule->cr.match, &match);
3761 fu.priority = rule->cr.priority;
3762 if (!(flags & NXFMF_ACTIONS)) {
3766 fu.ofpacts = rule->ofpacts;
3767 fu.ofpacts_len = rule->ofpacts_len;
3769 /* An operation is in progress. Use the previous version of the flow's
3770 * actions, so that when the operation commits we report the change. */
3772 case OFOPERATION_ADD:
3773 /* We already verified that there was a victim. */
3774 fu.ofpacts = op->victim->ofpacts;
3775 fu.ofpacts_len = op->victim->ofpacts_len;
3778 case OFOPERATION_MODIFY:
3780 fu.ofpacts = op->ofpacts;
3781 fu.ofpacts_len = op->ofpacts_len;
3783 fu.ofpacts = rule->ofpacts;
3784 fu.ofpacts_len = rule->ofpacts_len;
3788 case OFOPERATION_DELETE:
3789 fu.ofpacts = rule->ofpacts;
3790 fu.ofpacts_len = rule->ofpacts_len;
3798 if (list_is_empty(msgs)) {
3799 ofputil_start_flow_update(msgs);
3801 ofputil_append_flow_update(&fu, msgs);
3805 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3809 LIST_FOR_EACH (rule, ofproto_node, rules) {
3810 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3811 rule->monitor_flags = 0;
3813 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3818 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3819 struct rule *rule, uint64_t seqno,
3822 enum nx_flow_monitor_flags update;
3824 if (ofproto_rule_is_hidden(rule)) {
3829 ? ofoperation_has_out_port(rule->pending, m->out_port)
3830 : ofproto_rule_has_out_port(rule, m->out_port))) {
3835 if (rule->add_seqno > seqno) {
3836 update = NXFMF_ADD | NXFMF_MODIFY;
3837 } else if (rule->modify_seqno > seqno) {
3838 update = NXFMF_MODIFY;
3843 if (!(m->flags & update)) {
3847 update = NXFMF_INITIAL;
3850 if (!rule->monitor_flags) {
3851 list_push_back(rules, &rule->ofproto_node);
3853 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3857 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3861 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3862 const struct ofoperation *op;
3863 const struct oftable *table;
3864 struct cls_rule target;
3866 cls_rule_init_from_minimatch(&target, &m->match, 0);
3867 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3868 struct cls_cursor cursor;
3871 cls_cursor_init(&cursor, &table->cls, &target);
3872 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3873 ovs_assert(!rule->pending); /* XXX */
3874 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3878 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3879 struct rule *rule = op->rule;
3881 if (((m->table_id == 0xff
3882 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3883 : m->table_id == rule->table_id))
3884 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3885 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3888 cls_rule_destroy(&target);
3892 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3895 if (m->flags & NXFMF_INITIAL) {
3896 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3901 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3902 uint64_t seqno, struct list *rules)
3904 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3908 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3910 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3911 struct ofmonitor **monitors;
3912 size_t n_monitors, allocated_monitors;
3913 struct list replies;
3920 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3922 n_monitors = allocated_monitors = 0;
3924 struct ofputil_flow_monitor_request request;
3925 struct ofmonitor *m;
3928 retval = ofputil_decode_flow_monitor_request(&request, &b);
3929 if (retval == EOF) {
3931 } else if (retval) {
3936 if (request.table_id != 0xff
3937 && request.table_id >= ofproto->n_tables) {
3938 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3942 error = ofmonitor_create(&request, ofconn, &m);
3947 if (n_monitors >= allocated_monitors) {
3948 monitors = x2nrealloc(monitors, &allocated_monitors,
3951 monitors[n_monitors++] = m;
3955 for (i = 0; i < n_monitors; i++) {
3956 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3959 ofpmp_init(&replies, oh);
3960 ofmonitor_compose_refresh_updates(&rules, &replies);
3961 ofconn_send_replies(ofconn, &replies);
3968 for (i = 0; i < n_monitors; i++) {
3969 ofmonitor_destroy(monitors[i]);
3976 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3978 struct ofmonitor *m;
3981 id = ofputil_decode_flow_monitor_cancel(oh);
3982 m = ofmonitor_lookup(ofconn, id);
3984 return OFPERR_NXBRC_FM_BAD_ID;
3987 ofmonitor_destroy(m);
3992 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3994 const struct ofp_header *oh = msg->data;
3998 error = ofptype_decode(&type, oh);
4004 /* OpenFlow requests. */
4005 case OFPTYPE_ECHO_REQUEST:
4006 return handle_echo_request(ofconn, oh);
4008 case OFPTYPE_FEATURES_REQUEST:
4009 return handle_features_request(ofconn, oh);
4011 case OFPTYPE_GET_CONFIG_REQUEST:
4012 return handle_get_config_request(ofconn, oh);
4014 case OFPTYPE_SET_CONFIG:
4015 return handle_set_config(ofconn, oh);
4017 case OFPTYPE_PACKET_OUT:
4018 return handle_packet_out(ofconn, oh);
4020 case OFPTYPE_PORT_MOD:
4021 return handle_port_mod(ofconn, oh);
4023 case OFPTYPE_FLOW_MOD:
4024 return handle_flow_mod(ofconn, oh);
4026 case OFPTYPE_BARRIER_REQUEST:
4027 return handle_barrier_request(ofconn, oh);
4029 case OFPTYPE_ROLE_REQUEST:
4030 return handle_role_request(ofconn, oh);
4032 /* OpenFlow replies. */
4033 case OFPTYPE_ECHO_REPLY:
4036 /* Nicira extension requests. */
4037 case OFPTYPE_FLOW_MOD_TABLE_ID:
4038 return handle_nxt_flow_mod_table_id(ofconn, oh);
4040 case OFPTYPE_SET_FLOW_FORMAT:
4041 return handle_nxt_set_flow_format(ofconn, oh);
4043 case OFPTYPE_SET_PACKET_IN_FORMAT:
4044 return handle_nxt_set_packet_in_format(ofconn, oh);
4046 case OFPTYPE_SET_CONTROLLER_ID:
4047 return handle_nxt_set_controller_id(ofconn, oh);
4049 case OFPTYPE_FLOW_AGE:
4050 /* Nothing to do. */
4053 case OFPTYPE_FLOW_MONITOR_CANCEL:
4054 return handle_flow_monitor_cancel(ofconn, oh);
4056 case OFPTYPE_SET_ASYNC_CONFIG:
4057 return handle_nxt_set_async_config(ofconn, oh);
4059 /* Statistics requests. */
4060 case OFPTYPE_DESC_STATS_REQUEST:
4061 return handle_desc_stats_request(ofconn, oh);
4063 case OFPTYPE_FLOW_STATS_REQUEST:
4064 return handle_flow_stats_request(ofconn, oh);
4066 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4067 return handle_aggregate_stats_request(ofconn, oh);
4069 case OFPTYPE_TABLE_STATS_REQUEST:
4070 return handle_table_stats_request(ofconn, oh);
4072 case OFPTYPE_PORT_STATS_REQUEST:
4073 return handle_port_stats_request(ofconn, oh);
4075 case OFPTYPE_QUEUE_STATS_REQUEST:
4076 return handle_queue_stats_request(ofconn, oh);
4078 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4079 return handle_port_desc_stats_request(ofconn, oh);
4081 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4082 return handle_flow_monitor_request(ofconn, oh);
4084 /* FIXME: Change the following once they are implemented: */
4085 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4086 case OFPTYPE_GET_ASYNC_REQUEST:
4087 case OFPTYPE_METER_MOD:
4088 case OFPTYPE_GROUP_REQUEST:
4089 case OFPTYPE_GROUP_DESC_REQUEST:
4090 case OFPTYPE_GROUP_FEATURES_REQUEST:
4091 case OFPTYPE_METER_REQUEST:
4092 case OFPTYPE_METER_CONFIG_REQUEST:
4093 case OFPTYPE_METER_FEATURES_REQUEST:
4094 case OFPTYPE_TABLE_FEATURES_REQUEST:
4095 return OFPERR_OFPBRC_BAD_TYPE;
4099 case OFPTYPE_FEATURES_REPLY:
4100 case OFPTYPE_GET_CONFIG_REPLY:
4101 case OFPTYPE_PACKET_IN:
4102 case OFPTYPE_FLOW_REMOVED:
4103 case OFPTYPE_PORT_STATUS:
4104 case OFPTYPE_BARRIER_REPLY:
4105 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4106 case OFPTYPE_DESC_STATS_REPLY:
4107 case OFPTYPE_FLOW_STATS_REPLY:
4108 case OFPTYPE_QUEUE_STATS_REPLY:
4109 case OFPTYPE_PORT_STATS_REPLY:
4110 case OFPTYPE_TABLE_STATS_REPLY:
4111 case OFPTYPE_AGGREGATE_STATS_REPLY:
4112 case OFPTYPE_PORT_DESC_STATS_REPLY:
4113 case OFPTYPE_ROLE_REPLY:
4114 case OFPTYPE_FLOW_MONITOR_PAUSED:
4115 case OFPTYPE_FLOW_MONITOR_RESUMED:
4116 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4117 case OFPTYPE_GET_ASYNC_REPLY:
4118 case OFPTYPE_GROUP_REPLY:
4119 case OFPTYPE_GROUP_DESC_REPLY:
4120 case OFPTYPE_GROUP_FEATURES_REPLY:
4121 case OFPTYPE_METER_REPLY:
4122 case OFPTYPE_METER_CONFIG_REPLY:
4123 case OFPTYPE_METER_FEATURES_REPLY:
4124 case OFPTYPE_TABLE_FEATURES_REPLY:
4126 return OFPERR_OFPBRC_BAD_TYPE;
4131 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4133 int error = handle_openflow__(ofconn, ofp_msg);
4134 if (error && error != OFPROTO_POSTPONE) {
4135 ofconn_send_error(ofconn, ofp_msg->data, error);
4137 COVERAGE_INC(ofproto_recv_openflow);
4138 return error != OFPROTO_POSTPONE;
4141 /* Asynchronous operations. */
4143 /* Creates and returns a new ofopgroup that is not associated with any
4144 * OpenFlow connection.
4146 * The caller should add operations to the returned group with
4147 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4148 static struct ofopgroup *
4149 ofopgroup_create_unattached(struct ofproto *ofproto)
4151 struct ofopgroup *group = xzalloc(sizeof *group);
4152 group->ofproto = ofproto;
4153 list_init(&group->ofproto_node);
4154 list_init(&group->ops);
4155 list_init(&group->ofconn_node);
4159 /* Creates and returns a new ofopgroup for 'ofproto'.
4161 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4162 * connection. The 'request' and 'buffer_id' arguments are ignored.
4164 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4165 * If the ofopgroup eventually fails, then the error reply will include
4166 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4167 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4169 * The caller should add operations to the returned group with
4170 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4171 static struct ofopgroup *
4172 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4173 const struct ofp_header *request, uint32_t buffer_id)
4175 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4177 size_t request_len = ntohs(request->length);
4179 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4181 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4182 group->ofconn = ofconn;
4183 group->request = xmemdup(request, MIN(request_len, 64));
4184 group->buffer_id = buffer_id;
4189 /* Submits 'group' for processing.
4191 * If 'group' contains no operations (e.g. none were ever added, or all of the
4192 * ones that were added completed synchronously), then it is destroyed
4193 * immediately. Otherwise it is added to the ofproto's list of pending
4196 ofopgroup_submit(struct ofopgroup *group)
4198 if (!group->n_running) {
4199 ofopgroup_complete(group);
4201 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4202 group->ofproto->n_pending++;
4207 ofopgroup_complete(struct ofopgroup *group)
4209 struct ofproto *ofproto = group->ofproto;
4211 struct ofconn *abbrev_ofconn;
4212 ovs_be32 abbrev_xid;
4214 struct ofoperation *op, *next_op;
4217 ovs_assert(!group->n_running);
4220 LIST_FOR_EACH (op, group_node, &group->ops) {
4227 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4228 LIST_FOR_EACH (op, group_node, &group->ops) {
4229 if (op->type != OFOPERATION_DELETE) {
4230 struct ofpbuf *packet;
4233 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4237 error = rule_execute(op->rule, in_port, packet);
4244 if (!error && !list_is_empty(&group->ofconn_node)) {
4245 abbrev_ofconn = group->ofconn;
4246 abbrev_xid = group->request->xid;
4248 abbrev_ofconn = NULL;
4249 abbrev_xid = htonl(0);
4251 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4252 struct rule *rule = op->rule;
4254 /* We generally want to report the change to active OpenFlow flow
4255 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4257 - The operation failed.
4259 - The affected rule is not visible to controllers.
4261 - The operation's only effect was to update rule->modified. */
4263 || ofproto_rule_is_hidden(rule)
4264 || (op->type == OFOPERATION_MODIFY
4266 && rule->flow_cookie == op->flow_cookie))) {
4267 /* Check that we can just cast from ofoperation_type to
4268 * nx_flow_update_event. */
4269 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4271 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4273 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4276 ofmonitor_report(ofproto->connmgr, rule,
4277 (enum nx_flow_update_event) op->type,
4278 op->reason, abbrev_ofconn, abbrev_xid);
4281 rule->pending = NULL;
4284 case OFOPERATION_ADD:
4288 ofproto_rule_destroy__(op->victim);
4289 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4290 if (vid_mask == VLAN_VID_MASK) {
4291 if (ofproto->vlan_bitmap) {
4292 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4293 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4294 bitmap_set1(ofproto->vlan_bitmap, vid);
4295 ofproto->vlans_changed = true;
4298 ofproto->vlans_changed = true;
4302 oftable_substitute_rule(rule, op->victim);
4303 ofproto_rule_destroy__(rule);
4307 case OFOPERATION_DELETE:
4308 ovs_assert(!op->error);
4309 ofproto_rule_destroy__(rule);
4313 case OFOPERATION_MODIFY:
4315 rule->modified = time_msec();
4317 rule->flow_cookie = op->flow_cookie;
4319 free(rule->ofpacts);
4320 rule->ofpacts = op->ofpacts;
4321 rule->ofpacts_len = op->ofpacts_len;
4323 op->ofpacts_len = 0;
4332 ofoperation_destroy(op);
4335 ofmonitor_flush(ofproto->connmgr);
4337 if (!list_is_empty(&group->ofproto_node)) {
4338 ovs_assert(ofproto->n_pending > 0);
4339 ofproto->n_pending--;
4340 list_remove(&group->ofproto_node);
4342 if (!list_is_empty(&group->ofconn_node)) {
4343 list_remove(&group->ofconn_node);
4345 ofconn_send_error(group->ofconn, group->request, error);
4347 connmgr_retry(ofproto->connmgr);
4349 free(group->request);
4353 /* Initiates a new operation on 'rule', of the specified 'type', within
4354 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4356 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4357 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4359 * Returns the newly created ofoperation (which is also available as
4360 * rule->pending). */
4361 static struct ofoperation *
4362 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4363 enum ofoperation_type type,
4364 enum ofp_flow_removed_reason reason)
4366 struct ofproto *ofproto = group->ofproto;
4367 struct ofoperation *op;
4369 ovs_assert(!rule->pending);
4371 op = rule->pending = xzalloc(sizeof *op);
4373 list_push_back(&group->ops, &op->group_node);
4376 op->reason = reason;
4377 op->flow_cookie = rule->flow_cookie;
4381 if (type == OFOPERATION_DELETE) {
4382 hmap_insert(&ofproto->deletions, &op->hmap_node,
4383 cls_rule_hash(&rule->cr, rule->table_id));
4390 ofoperation_destroy(struct ofoperation *op)
4392 struct ofopgroup *group = op->group;
4395 op->rule->pending = NULL;
4397 if (op->type == OFOPERATION_DELETE) {
4398 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4400 list_remove(&op->group_node);
4405 /* Indicates that 'op' completed with status 'error', which is either 0 to
4406 * indicate success or an OpenFlow error code on failure.
4408 * If 'error' is 0, indicating success, the operation will be committed
4409 * permanently to the flow table. There is one interesting subcase:
4411 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4412 * the flow table (the "victim" rule) by a new one, then the caller must
4413 * have uninitialized any derived state in the victim rule, as in step 5 in
4414 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4415 * performs steps 6 and 7 for the victim rule, most notably by calling its
4416 * ->rule_dealloc() function.
4418 * If 'error' is nonzero, then generally the operation will be rolled back:
4420 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4421 * restores the original rule. The caller must have uninitialized any
4422 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4423 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4424 * and 7 for the new rule, calling its ->rule_dealloc() function.
4426 * - If 'op' is a "modify flow" operation, ofproto restores the original
4429 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4430 * allowed to fail. It must always succeed.
4432 * Please see the large comment in ofproto/ofproto-provider.h titled
4433 * "Asynchronous Operation Support" for more information. */
4435 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4437 struct ofopgroup *group = op->group;
4439 ovs_assert(op->rule->pending == op);
4440 ovs_assert(group->n_running > 0);
4441 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4444 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4445 ofopgroup_complete(group);
4450 ofoperation_get_victim(struct ofoperation *op)
4452 ovs_assert(op->type == OFOPERATION_ADD);
4457 pick_datapath_id(const struct ofproto *ofproto)
4459 const struct ofport *port;
4461 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4463 uint8_t ea[ETH_ADDR_LEN];
4466 error = netdev_get_etheraddr(port->netdev, ea);
4468 return eth_addr_to_uint64(ea);
4470 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4471 ofproto->name, netdev_get_name(port->netdev),
4474 return ofproto->fallback_dpid;
4478 pick_fallback_dpid(void)
4480 uint8_t ea[ETH_ADDR_LEN];
4481 eth_addr_nicira_random(ea);
4482 return eth_addr_to_uint64(ea);
4485 /* Table overflow policy. */
4487 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4488 * is not configured to evict rules or if the table contains no evictable
4489 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4491 static struct rule *
4492 choose_rule_to_evict(struct oftable *table)
4494 struct eviction_group *evg;
4496 if (!table->eviction_fields) {
4500 /* In the common case, the outer and inner loops here will each be entered
4503 * - The inner loop normally "return"s in its first iteration. If the
4504 * eviction group has any evictable rules, then it always returns in
4507 * - The outer loop only iterates more than once if the largest eviction
4508 * group has no evictable rules.
4510 * - The outer loop can exit only if table's 'max_flows' is all filled up
4511 * by unevictable rules'. */
4512 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4515 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4516 if (rule->evictable) {
4525 /* Searches 'ofproto' for tables that have more flows than their configured
4526 * maximum and that have flow eviction enabled, and evicts as many flows as
4527 * necessary and currently feasible from them.
4529 * This triggers only when an OpenFlow table has N flows in it and then the
4530 * client configures a maximum number of flows less than N. */
4532 ofproto_evict(struct ofproto *ofproto)
4534 struct ofopgroup *group;
4535 struct oftable *table;
4537 group = ofopgroup_create_unattached(ofproto);
4538 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4539 while (classifier_count(&table->cls) > table->max_flows
4540 && table->eviction_fields) {
4543 rule = choose_rule_to_evict(table);
4544 if (!rule || rule->pending) {
4548 ofoperation_create(group, rule,
4549 OFOPERATION_DELETE, OFPRR_EVICTION);
4550 oftable_remove_rule(rule);
4551 ofproto->ofproto_class->rule_destruct(rule);
4554 ofopgroup_submit(group);
4557 /* Eviction groups. */
4559 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4560 * rules. The priority contains low-order random bits to ensure that eviction
4561 * groups with the same number of rules are prioritized randomly. */
4563 eviction_group_priority(size_t n_rules)
4565 uint16_t size = MIN(UINT16_MAX, n_rules);
4566 return (size << 16) | random_uint16();
4569 /* Updates 'evg', an eviction_group within 'table', following a change that
4570 * adds or removes rules in 'evg'. */
4572 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4574 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4575 eviction_group_priority(heap_count(&evg->rules)));
4578 /* Destroys 'evg', an eviction_group within 'table':
4580 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4581 * rules themselves, just removes them from the eviction group.)
4583 * - Removes 'evg' from 'table'.
4587 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4589 while (!heap_is_empty(&evg->rules)) {
4592 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4593 rule->eviction_group = NULL;
4595 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4596 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4597 heap_destroy(&evg->rules);
4601 /* Removes 'rule' from its eviction group, if any. */
4603 eviction_group_remove_rule(struct rule *rule)
4605 if (rule->eviction_group) {
4606 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4607 struct eviction_group *evg = rule->eviction_group;
4609 rule->eviction_group = NULL;
4610 heap_remove(&evg->rules, &rule->evg_node);
4611 if (heap_is_empty(&evg->rules)) {
4612 eviction_group_destroy(table, evg);
4614 eviction_group_resized(table, evg);
4619 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4620 * returns the hash value. */
4622 eviction_group_hash_rule(struct rule *rule)
4624 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4625 const struct mf_subfield *sf;
4629 hash = table->eviction_group_id_basis;
4630 miniflow_expand(&rule->cr.match.flow, &flow);
4631 for (sf = table->eviction_fields;
4632 sf < &table->eviction_fields[table->n_eviction_fields];
4635 if (mf_are_prereqs_ok(sf->field, &flow)) {
4636 union mf_value value;
4638 mf_get_value(sf->field, &flow, &value);
4640 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4642 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4643 unsigned int start = sf->ofs + sf->n_bits;
4644 bitwise_zero(&value, sf->field->n_bytes, start,
4645 sf->field->n_bytes * 8 - start);
4647 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4649 hash = hash_int(hash, 0);
4656 /* Returns an eviction group within 'table' with the given 'id', creating one
4658 static struct eviction_group *
4659 eviction_group_find(struct oftable *table, uint32_t id)
4661 struct eviction_group *evg;
4663 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4667 evg = xmalloc(sizeof *evg);
4668 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4669 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4670 eviction_group_priority(0));
4671 heap_init(&evg->rules);
4676 /* Returns an eviction priority for 'rule'. The return value should be
4677 * interpreted so that higher priorities make a rule more attractive candidates
4680 rule_eviction_priority(struct rule *rule)
4682 long long int hard_expiration;
4683 long long int idle_expiration;
4684 long long int expiration;
4685 uint32_t expiration_offset;
4687 /* Calculate time of expiration. */
4688 hard_expiration = (rule->hard_timeout
4689 ? rule->modified + rule->hard_timeout * 1000
4691 idle_expiration = (rule->idle_timeout
4692 ? rule->used + rule->idle_timeout * 1000
4694 expiration = MIN(hard_expiration, idle_expiration);
4695 if (expiration == LLONG_MAX) {
4699 /* Calculate the time of expiration as a number of (approximate) seconds
4700 * after program startup.
4702 * This should work OK for program runs that last UINT32_MAX seconds or
4703 * less. Therefore, please restart OVS at least once every 136 years. */
4704 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4706 /* Invert the expiration offset because we're using a max-heap. */
4707 return UINT32_MAX - expiration_offset;
4710 /* Adds 'rule' to an appropriate eviction group for its oftable's
4711 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4712 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4715 * The caller must ensure that 'rule' is not already in an eviction group. */
4717 eviction_group_add_rule(struct rule *rule)
4719 struct ofproto *ofproto = rule->ofproto;
4720 struct oftable *table = &ofproto->tables[rule->table_id];
4722 if (table->eviction_fields
4723 && (rule->hard_timeout || rule->idle_timeout)) {
4724 struct eviction_group *evg;
4726 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4728 rule->eviction_group = evg;
4729 heap_insert(&evg->rules, &rule->evg_node,
4730 rule_eviction_priority(rule));
4731 eviction_group_resized(table, evg);
4737 /* Initializes 'table'. */
4739 oftable_init(struct oftable *table)
4741 memset(table, 0, sizeof *table);
4742 classifier_init(&table->cls);
4743 table->max_flows = UINT_MAX;
4746 /* Destroys 'table', including its classifier and eviction groups.
4748 * The caller is responsible for freeing 'table' itself. */
4750 oftable_destroy(struct oftable *table)
4752 ovs_assert(classifier_is_empty(&table->cls));
4753 oftable_disable_eviction(table);
4754 classifier_destroy(&table->cls);
4758 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4759 * string, then 'table' will use its default name.
4761 * This only affects the name exposed for a table exposed through the OpenFlow
4762 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4764 oftable_set_name(struct oftable *table, const char *name)
4766 if (name && name[0]) {
4767 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4768 if (!table->name || strncmp(name, table->name, len)) {
4770 table->name = xmemdup0(name, len);
4778 /* oftables support a choice of two policies when adding a rule would cause the
4779 * number of flows in the table to exceed the configured maximum number: either
4780 * they can refuse to add the new flow or they can evict some existing flow.
4781 * This function configures the former policy on 'table'. */
4783 oftable_disable_eviction(struct oftable *table)
4785 if (table->eviction_fields) {
4786 struct eviction_group *evg, *next;
4788 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4789 &table->eviction_groups_by_id) {
4790 eviction_group_destroy(table, evg);
4792 hmap_destroy(&table->eviction_groups_by_id);
4793 heap_destroy(&table->eviction_groups_by_size);
4795 free(table->eviction_fields);
4796 table->eviction_fields = NULL;
4797 table->n_eviction_fields = 0;
4801 /* oftables support a choice of two policies when adding a rule would cause the
4802 * number of flows in the table to exceed the configured maximum number: either
4803 * they can refuse to add the new flow or they can evict some existing flow.
4804 * This function configures the latter policy on 'table', with fairness based
4805 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4806 * 'n_fields' as 0 disables fairness.) */
4808 oftable_enable_eviction(struct oftable *table,
4809 const struct mf_subfield *fields, size_t n_fields)
4811 struct cls_cursor cursor;
4814 if (table->eviction_fields
4815 && n_fields == table->n_eviction_fields
4817 || !memcmp(fields, table->eviction_fields,
4818 n_fields * sizeof *fields))) {
4823 oftable_disable_eviction(table);
4825 table->n_eviction_fields = n_fields;
4826 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4828 table->eviction_group_id_basis = random_uint32();
4829 hmap_init(&table->eviction_groups_by_id);
4830 heap_init(&table->eviction_groups_by_size);
4832 cls_cursor_init(&cursor, &table->cls, NULL);
4833 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4834 eviction_group_add_rule(rule);
4838 /* Removes 'rule' from the oftable that contains it. */
4840 oftable_remove_rule(struct rule *rule)
4842 struct ofproto *ofproto = rule->ofproto;
4843 struct oftable *table = &ofproto->tables[rule->table_id];
4845 classifier_remove(&table->cls, &rule->cr);
4846 eviction_group_remove_rule(rule);
4847 if (!list_is_empty(&rule->expirable)) {
4848 list_remove(&rule->expirable);
4852 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4853 * oftable that has an identical cls_rule. Returns the rule that was removed,
4854 * if any, and otherwise NULL. */
4855 static struct rule *
4856 oftable_replace_rule(struct rule *rule)
4858 struct ofproto *ofproto = rule->ofproto;
4859 struct oftable *table = &ofproto->tables[rule->table_id];
4860 struct rule *victim;
4861 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4864 list_insert(&ofproto->expirable, &rule->expirable);
4867 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4869 if (!list_is_empty(&victim->expirable)) {
4870 list_remove(&victim->expirable);
4872 eviction_group_remove_rule(victim);
4874 eviction_group_add_rule(rule);
4878 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4880 oftable_substitute_rule(struct rule *old, struct rule *new)
4883 oftable_replace_rule(new);
4885 oftable_remove_rule(old);
4889 /* unixctl commands. */
4892 ofproto_lookup(const char *name)
4894 struct ofproto *ofproto;
4896 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4898 if (!strcmp(ofproto->name, name)) {
4906 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4907 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4909 struct ofproto *ofproto;
4913 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4914 ds_put_format(&results, "%s\n", ofproto->name);
4916 unixctl_command_reply(conn, ds_cstr(&results));
4917 ds_destroy(&results);
4921 ofproto_unixctl_init(void)
4923 static bool registered;
4929 unixctl_command_register("ofproto/list", "", 0, 0,
4930 ofproto_unixctl_list, NULL);
4933 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4935 * This is deprecated. It is only for compatibility with broken device drivers
4936 * in old versions of Linux that do not properly support VLANs when VLAN
4937 * devices are not used. When broken device drivers are no longer in
4938 * widespread use, we will delete these interfaces. */
4940 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4941 * (exactly) by an OpenFlow rule in 'ofproto'. */
4943 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4945 const struct oftable *oftable;
4947 free(ofproto->vlan_bitmap);
4948 ofproto->vlan_bitmap = bitmap_allocate(4096);
4949 ofproto->vlans_changed = false;
4951 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4952 const struct cls_table *table;
4954 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4955 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4956 const struct cls_rule *rule;
4958 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4959 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4960 bitmap_set1(vlan_bitmap, vid);
4961 bitmap_set1(ofproto->vlan_bitmap, vid);
4968 /* Returns true if new VLANs have come into use by the flow table since the
4969 * last call to ofproto_get_vlan_usage().
4971 * We don't track when old VLANs stop being used. */
4973 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4975 return ofproto->vlans_changed;
4978 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4979 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4980 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4981 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4982 * then the VLAN device is un-enslaved. */
4984 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4985 uint16_t realdev_ofp_port, int vid)
4987 struct ofport *ofport;
4990 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
4992 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4994 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4995 ofproto->name, vlandev_ofp_port);
4999 if (!ofproto->ofproto_class->set_realdev) {
5000 if (!vlandev_ofp_port) {
5003 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5007 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5009 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5010 ofproto->name, vlandev_ofp_port,
5011 netdev_get_name(ofport->netdev), strerror(error));