2 * Copyright (c) 2009, 2010, 2011, 2012 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 = xstrdup(DEFAULT_MFR_DESC);
411 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
412 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
413 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
414 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
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 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
423 ofproto->state = S_OPENFLOW;
424 list_init(&ofproto->pending);
425 ofproto->n_pending = 0;
426 hmap_init(&ofproto->deletions);
427 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
428 ofproto->first_op = ofproto->last_op = LLONG_MIN;
429 ofproto->next_op_report = LLONG_MAX;
430 ofproto->op_backoff = LLONG_MIN;
431 ofproto->vlan_bitmap = NULL;
432 ofproto->vlans_changed = false;
433 ofproto->min_mtu = INT_MAX;
435 error = ofproto->ofproto_class->construct(ofproto);
437 VLOG_ERR("failed to open datapath %s: %s",
438 datapath_name, strerror(error));
439 ofproto_destroy__(ofproto);
443 /* The "max_ports" member should have been set by ->construct(ofproto).
444 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
445 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
446 bitmap_set1(ofproto->ofp_port_ids, 0);
448 /* Check that hidden tables, if any, are at the end. */
449 assert(ofproto->n_tables);
450 for (i = 0; i + 1 < ofproto->n_tables; i++) {
451 enum oftable_flags flags = ofproto->tables[i].flags;
452 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
454 assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
457 ofproto->datapath_id = pick_datapath_id(ofproto);
464 /* Must be called (only) by an ofproto implementation in its constructor
465 * function. See the large comment on 'construct' in struct ofproto_class for
468 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
470 struct oftable *table;
472 assert(!ofproto->n_tables);
473 assert(n_tables >= 1 && n_tables <= 255);
475 ofproto->n_tables = n_tables;
476 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
477 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
482 /* To be optionally called (only) by an ofproto implementation in its
483 * constructor function. See the large comment on 'construct' in struct
484 * ofproto_class for details.
486 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
487 * will then ensure that actions passed into the ofproto implementation will
488 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
489 * function is not called, there will be no such restriction.
491 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
492 * the 'max_ports' restriction. */
494 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
496 assert(max_ports <= OFPP_MAX);
497 ofproto->max_ports = max_ports;
501 ofproto_get_datapath_id(const struct ofproto *ofproto)
503 return ofproto->datapath_id;
507 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
509 uint64_t old_dpid = p->datapath_id;
510 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
511 if (p->datapath_id != old_dpid) {
512 /* Force all active connections to reconnect, since there is no way to
513 * notify a controller that the datapath ID has changed. */
514 ofproto_reconnect_controllers(p);
519 ofproto_set_controllers(struct ofproto *p,
520 const struct ofproto_controller *controllers,
521 size_t n_controllers, uint32_t allowed_versions)
523 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
528 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
530 connmgr_set_fail_mode(p->connmgr, fail_mode);
533 /* Drops the connections between 'ofproto' and all of its controllers, forcing
534 * them to reconnect. */
536 ofproto_reconnect_controllers(struct ofproto *ofproto)
538 connmgr_reconnect(ofproto->connmgr);
541 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
542 * in-band control should guarantee access, in the same way that in-band
543 * control guarantees access to OpenFlow controllers. */
545 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
546 const struct sockaddr_in *extras, size_t n)
548 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
551 /* Sets the OpenFlow queue used by flows set up by in-band control on
552 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
553 * flows will use the default queue. */
555 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
557 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
560 /* Sets the number of flows at which eviction from the kernel flow table
563 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
565 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
566 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
568 ofproto->flow_eviction_threshold = threshold;
572 /* If forward_bpdu is true, the NORMAL action will forward frames with
573 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
574 * the NORMAL action will drop these frames. */
576 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
578 bool old_val = ofproto->forward_bpdu;
579 ofproto->forward_bpdu = forward_bpdu;
580 if (old_val != ofproto->forward_bpdu) {
581 if (ofproto->ofproto_class->forward_bpdu_changed) {
582 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
587 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
588 * 'idle_time', in seconds, and the maximum number of MAC table entries to
591 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
594 if (ofproto->ofproto_class->set_mac_table_config) {
595 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
601 ofproto_set_desc(struct ofproto *p,
602 const char *mfr_desc, const char *hw_desc,
603 const char *sw_desc, const char *serial_desc,
606 struct ofp_desc_stats *ods;
609 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
610 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
611 p->name, sizeof ods->mfr_desc);
614 p->mfr_desc = xstrdup(mfr_desc);
617 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
618 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
619 p->name, sizeof ods->hw_desc);
622 p->hw_desc = xstrdup(hw_desc);
625 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
626 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
627 p->name, sizeof ods->sw_desc);
630 p->sw_desc = xstrdup(sw_desc);
633 if (strlen(serial_desc) >= sizeof ods->serial_num) {
634 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
635 "bytes", p->name, sizeof ods->serial_num);
637 free(p->serial_desc);
638 p->serial_desc = xstrdup(serial_desc);
641 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
642 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
643 p->name, sizeof ods->dp_desc);
646 p->dp_desc = xstrdup(dp_desc);
651 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
653 return connmgr_set_snoops(ofproto->connmgr, snoops);
657 ofproto_set_netflow(struct ofproto *ofproto,
658 const struct netflow_options *nf_options)
660 if (nf_options && sset_is_empty(&nf_options->collectors)) {
664 if (ofproto->ofproto_class->set_netflow) {
665 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
667 return nf_options ? EOPNOTSUPP : 0;
672 ofproto_set_sflow(struct ofproto *ofproto,
673 const struct ofproto_sflow_options *oso)
675 if (oso && sset_is_empty(&oso->targets)) {
679 if (ofproto->ofproto_class->set_sflow) {
680 return ofproto->ofproto_class->set_sflow(ofproto, oso);
682 return oso ? EOPNOTSUPP : 0;
686 /* Spanning Tree Protocol (STP) configuration. */
688 /* Configures STP on 'ofproto' using the settings defined in 's'. If
689 * 's' is NULL, disables STP.
691 * Returns 0 if successful, otherwise a positive errno value. */
693 ofproto_set_stp(struct ofproto *ofproto,
694 const struct ofproto_stp_settings *s)
696 return (ofproto->ofproto_class->set_stp
697 ? ofproto->ofproto_class->set_stp(ofproto, s)
701 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
702 * 'enabled' member of 's' is false, then the other members are not
705 * Returns 0 if successful, otherwise a positive errno value. */
707 ofproto_get_stp_status(struct ofproto *ofproto,
708 struct ofproto_stp_status *s)
710 return (ofproto->ofproto_class->get_stp_status
711 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
715 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
716 * in 's'. The caller is responsible for assigning STP port numbers
717 * (using the 'port_num' member in the range of 1 through 255, inclusive)
718 * and ensuring there are no duplicates. If the 's' is NULL, then STP
719 * is disabled on the port.
721 * Returns 0 if successful, otherwise a positive errno value.*/
723 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
724 const struct ofproto_port_stp_settings *s)
726 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
728 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
729 ofproto->name, ofp_port);
733 return (ofproto->ofproto_class->set_stp_port
734 ? ofproto->ofproto_class->set_stp_port(ofport, s)
738 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
739 * 's'. If the 'enabled' member in 's' is false, then the other members
740 * are not meaningful.
742 * Returns 0 if successful, otherwise a positive errno value.*/
744 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
745 struct ofproto_port_stp_status *s)
747 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
749 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
750 "port %"PRIu16, ofproto->name, ofp_port);
754 return (ofproto->ofproto_class->get_stp_port_status
755 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
759 /* Queue DSCP configuration. */
761 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
762 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
763 * to implement QoS. Instead, it is used to implement features which require
764 * knowledge of what queues exist on a port, and some basic information about
767 * Returns 0 if successful, otherwise a positive errno value. */
769 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
770 const struct ofproto_port_queue *queues,
773 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
776 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
777 ofproto->name, ofp_port);
781 return (ofproto->ofproto_class->set_queues
782 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
786 /* Connectivity Fault Management configuration. */
788 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
790 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
792 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
793 if (ofport && ofproto->ofproto_class->set_cfm) {
794 ofproto->ofproto_class->set_cfm(ofport, NULL);
798 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
799 * basic configuration from the configuration members in 'cfm', and the remote
800 * maintenance point ID from remote_mpid. Ignores the statistics members of
803 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
805 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
806 const struct cfm_settings *s)
808 struct ofport *ofport;
811 ofport = ofproto_get_port(ofproto, ofp_port);
813 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
814 ofproto->name, ofp_port);
818 /* XXX: For configuration simplicity, we only support one remote_mpid
819 * outside of the CFM module. It's not clear if this is the correct long
820 * term solution or not. */
821 error = (ofproto->ofproto_class->set_cfm
822 ? ofproto->ofproto_class->set_cfm(ofport, s)
825 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
826 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
831 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
832 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
833 * 0 if LACP partner information is not current (generally indicating a
834 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
836 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
838 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
839 return (ofport && ofproto->ofproto_class->port_is_lacp_current
840 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
846 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
847 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
848 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
849 * configuration plus, if there is more than one slave, a bonding
852 * If 'aux' is already registered then this function updates its configuration
853 * to 's'. Otherwise, this function registers a new bundle.
855 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
858 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
859 const struct ofproto_bundle_settings *s)
861 return (ofproto->ofproto_class->bundle_set
862 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
866 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
867 * If no such bundle has been registered, this has no effect. */
869 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
871 return ofproto_bundle_register(ofproto, aux, NULL);
875 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
876 * If 'aux' is already registered then this function updates its configuration
877 * to 's'. Otherwise, this function registers a new mirror. */
879 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
880 const struct ofproto_mirror_settings *s)
882 return (ofproto->ofproto_class->mirror_set
883 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
887 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
888 * If no mirror has been registered, this has no effect. */
890 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
892 return ofproto_mirror_register(ofproto, aux, NULL);
895 /* Retrieves statistics from mirror associated with client data pointer
896 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
897 * 'bytes', respectively. If a particular counters is not supported,
898 * the appropriate argument is set to UINT64_MAX. */
900 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
901 uint64_t *packets, uint64_t *bytes)
903 if (!ofproto->ofproto_class->mirror_get_stats) {
904 *packets = *bytes = UINT64_MAX;
908 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
912 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
913 * which all packets are flooded, instead of using MAC learning. If
914 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
916 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
919 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
921 return (ofproto->ofproto_class->set_flood_vlans
922 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
926 /* Returns true if 'aux' is a registered bundle that is currently in use as the
927 * output for a mirror. */
929 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
931 return (ofproto->ofproto_class->is_mirror_output_bundle
932 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
936 /* Configuration of OpenFlow tables. */
938 /* Returns the number of OpenFlow tables in 'ofproto'. */
940 ofproto_get_n_tables(const struct ofproto *ofproto)
942 return ofproto->n_tables;
945 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
946 * settings from 's'. 'table_id' must be in the range 0 through the number of
947 * OpenFlow tables in 'ofproto' minus 1, inclusive.
949 * For read-only tables, only the name may be configured. */
951 ofproto_configure_table(struct ofproto *ofproto, int table_id,
952 const struct ofproto_table_settings *s)
954 struct oftable *table;
956 assert(table_id >= 0 && table_id < ofproto->n_tables);
957 table = &ofproto->tables[table_id];
959 oftable_set_name(table, s->name);
961 if (table->flags & OFTABLE_READONLY) {
966 oftable_enable_eviction(table, s->groups, s->n_groups);
968 oftable_disable_eviction(table);
971 table->max_flows = s->max_flows;
972 if (classifier_count(&table->cls) > table->max_flows
973 && table->eviction_fields) {
974 /* 'table' contains more flows than allowed. We might not be able to
975 * evict them right away because of the asynchronous nature of flow
976 * table changes. Schedule eviction for later. */
977 switch (ofproto->state) {
979 ofproto->state = S_EVICT;
983 /* We're already deleting flows, nothing more to do. */
990 ofproto_has_snoops(const struct ofproto *ofproto)
992 return connmgr_has_snoops(ofproto->connmgr);
996 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
998 connmgr_get_snoops(ofproto->connmgr, snoops);
1002 ofproto_flush__(struct ofproto *ofproto)
1004 struct ofopgroup *group;
1005 struct oftable *table;
1007 if (ofproto->ofproto_class->flush) {
1008 ofproto->ofproto_class->flush(ofproto);
1011 group = ofopgroup_create_unattached(ofproto);
1012 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1013 struct rule *rule, *next_rule;
1014 struct cls_cursor cursor;
1016 if (table->flags & OFTABLE_HIDDEN) {
1020 cls_cursor_init(&cursor, &table->cls, NULL);
1021 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1022 if (!rule->pending) {
1023 ofoperation_create(group, rule, OFOPERATION_DELETE,
1025 oftable_remove_rule(rule);
1026 ofproto->ofproto_class->rule_destruct(rule);
1030 ofopgroup_submit(group);
1034 ofproto_destroy__(struct ofproto *ofproto)
1036 struct oftable *table;
1038 assert(list_is_empty(&ofproto->pending));
1039 assert(!ofproto->n_pending);
1041 connmgr_destroy(ofproto->connmgr);
1043 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1044 free(ofproto->name);
1045 free(ofproto->type);
1046 free(ofproto->mfr_desc);
1047 free(ofproto->hw_desc);
1048 free(ofproto->sw_desc);
1049 free(ofproto->serial_desc);
1050 free(ofproto->dp_desc);
1051 hmap_destroy(&ofproto->ports);
1052 shash_destroy(&ofproto->port_by_name);
1053 bitmap_free(ofproto->ofp_port_ids);
1054 simap_destroy(&ofproto->ofp_requests);
1056 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1057 oftable_destroy(table);
1059 free(ofproto->tables);
1061 hmap_destroy(&ofproto->deletions);
1063 free(ofproto->vlan_bitmap);
1065 ofproto->ofproto_class->dealloc(ofproto);
1069 ofproto_destroy(struct ofproto *p)
1071 struct ofport *ofport, *next_ofport;
1078 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1079 ofport_destroy(ofport);
1082 p->ofproto_class->destruct(p);
1083 ofproto_destroy__(p);
1086 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1087 * kernel datapath, for example, this destroys the datapath in the kernel, and
1088 * with the netdev-based datapath, it tears down the data structures that
1089 * represent the datapath.
1091 * The datapath should not be currently open as an ofproto. */
1093 ofproto_delete(const char *name, const char *type)
1095 const struct ofproto_class *class = ofproto_class_find__(type);
1096 return (!class ? EAFNOSUPPORT
1097 : !class->del ? EACCES
1098 : class->del(type, name));
1102 process_port_change(struct ofproto *ofproto, int error, char *devname)
1104 if (error == ENOBUFS) {
1105 reinit_ports(ofproto);
1106 } else if (!error) {
1107 update_port(ofproto, devname);
1113 ofproto_type_run(const char *datapath_type)
1115 const struct ofproto_class *class;
1118 datapath_type = ofproto_normalize_type(datapath_type);
1119 class = ofproto_class_find__(datapath_type);
1121 error = class->type_run ? class->type_run(datapath_type) : 0;
1122 if (error && error != EAGAIN) {
1123 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1124 datapath_type, strerror(error));
1130 ofproto_type_run_fast(const char *datapath_type)
1132 const struct ofproto_class *class;
1135 datapath_type = ofproto_normalize_type(datapath_type);
1136 class = ofproto_class_find__(datapath_type);
1138 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1139 if (error && error != EAGAIN) {
1140 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1141 datapath_type, strerror(error));
1147 ofproto_type_wait(const char *datapath_type)
1149 const struct ofproto_class *class;
1151 datapath_type = ofproto_normalize_type(datapath_type);
1152 class = ofproto_class_find__(datapath_type);
1154 if (class->type_wait) {
1155 class->type_wait(datapath_type);
1160 ofproto_run(struct ofproto *p)
1162 struct sset changed_netdevs;
1163 const char *changed_netdev;
1164 struct ofport *ofport;
1167 error = p->ofproto_class->run(p);
1168 if (error && error != EAGAIN) {
1169 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1172 if (p->ofproto_class->port_poll) {
1175 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1176 process_port_change(p, error, devname);
1180 /* Update OpenFlow port status for any port whose netdev has changed.
1182 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1183 * destroyed, so it's not safe to update ports directly from the
1184 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1185 * need this two-phase approach. */
1186 sset_init(&changed_netdevs);
1187 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1188 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1189 if (ofport->change_seq != change_seq) {
1190 ofport->change_seq = change_seq;
1191 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1194 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1195 update_port(p, changed_netdev);
1197 sset_destroy(&changed_netdevs);
1201 connmgr_run(p->connmgr, handle_openflow);
1205 connmgr_run(p->connmgr, NULL);
1207 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1208 p->state = S_OPENFLOW;
1213 connmgr_run(p->connmgr, NULL);
1215 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1216 connmgr_flushed(p->connmgr);
1217 p->state = S_OPENFLOW;
1225 if (time_msec() >= p->next_op_report) {
1226 long long int ago = (time_msec() - p->first_op) / 1000;
1227 long long int interval = (p->last_op - p->first_op) / 1000;
1231 ds_put_format(&s, "%d flow_mods ",
1232 p->n_add + p->n_delete + p->n_modify);
1233 if (interval == ago) {
1234 ds_put_format(&s, "in the last %lld s", ago);
1235 } else if (interval) {
1236 ds_put_format(&s, "in the %lld s starting %lld s ago",
1239 ds_put_format(&s, "%lld s ago", ago);
1242 ds_put_cstr(&s, " (");
1244 ds_put_format(&s, "%d adds, ", p->n_add);
1247 ds_put_format(&s, "%d deletes, ", p->n_delete);
1250 ds_put_format(&s, "%d modifications, ", p->n_modify);
1253 ds_put_char(&s, ')');
1255 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1258 p->n_add = p->n_delete = p->n_modify = 0;
1259 p->next_op_report = LLONG_MAX;
1265 /* Performs periodic activity required by 'ofproto' that needs to be done
1266 * with the least possible latency.
1268 * It makes sense to call this function a couple of times per poll loop, to
1269 * provide a significant performance boost on some benchmarks with the
1270 * ofproto-dpif implementation. */
1272 ofproto_run_fast(struct ofproto *p)
1276 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1277 if (error && error != EAGAIN) {
1278 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1279 p->name, strerror(error));
1285 ofproto_wait(struct ofproto *p)
1287 struct ofport *ofport;
1289 p->ofproto_class->wait(p);
1290 if (p->ofproto_class->port_poll_wait) {
1291 p->ofproto_class->port_poll_wait(p);
1294 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1295 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1296 poll_immediate_wake();
1302 connmgr_wait(p->connmgr, true);
1307 connmgr_wait(p->connmgr, false);
1308 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1309 poll_immediate_wake();
1316 ofproto_is_alive(const struct ofproto *p)
1318 return connmgr_has_controllers(p->connmgr);
1321 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1322 * memory_report(). */
1324 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1326 const struct oftable *table;
1327 unsigned int n_rules;
1329 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1330 simap_increase(usage, "ops",
1331 ofproto->n_pending + hmap_count(&ofproto->deletions));
1334 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1335 n_rules += classifier_count(&table->cls);
1337 simap_increase(usage, "rules", n_rules);
1339 if (ofproto->ofproto_class->get_memory_usage) {
1340 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1343 connmgr_get_memory_usage(ofproto->connmgr, usage);
1347 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1350 connmgr_get_controller_info(ofproto->connmgr, info);
1354 ofproto_free_ofproto_controller_info(struct shash *info)
1356 connmgr_free_controller_info(info);
1359 /* Makes a deep copy of 'old' into 'port'. */
1361 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1363 port->name = xstrdup(old->name);
1364 port->type = xstrdup(old->type);
1365 port->ofp_port = old->ofp_port;
1368 /* Frees memory allocated to members of 'ofproto_port'.
1370 * Do not call this function on an ofproto_port obtained from
1371 * ofproto_port_dump_next(): that function retains ownership of the data in the
1374 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1376 free(ofproto_port->name);
1377 free(ofproto_port->type);
1380 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1382 * This function provides no status indication. An error status for the entire
1383 * dump operation is provided when it is completed by calling
1384 * ofproto_port_dump_done().
1387 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1388 const struct ofproto *ofproto)
1390 dump->ofproto = ofproto;
1391 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1395 /* Attempts to retrieve another port from 'dump', which must have been created
1396 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1397 * 'port' and returns true. On failure, returns false.
1399 * Failure might indicate an actual error or merely that the last port has been
1400 * dumped. An error status for the entire dump operation is provided when it
1401 * is completed by calling ofproto_port_dump_done().
1403 * The ofproto owns the data stored in 'port'. It will remain valid until at
1404 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1405 * ofproto_port_dump_done(). */
1407 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1408 struct ofproto_port *port)
1410 const struct ofproto *ofproto = dump->ofproto;
1416 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1419 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1425 /* Completes port table dump operation 'dump', which must have been created
1426 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1427 * error-free, otherwise a positive errno value describing the problem. */
1429 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1431 const struct ofproto *ofproto = dump->ofproto;
1433 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1436 return dump->error == EOF ? 0 : dump->error;
1439 /* Returns the type to pass to netdev_open() when a datapath of type
1440 * 'datapath_type' has a port of type 'port_type', for a few special
1441 * cases when a netdev type differs from a port type. For example, when
1442 * using the userspace datapath, a port of type "internal" needs to be
1445 * Returns either 'type' itself or a string literal, which must not be
1448 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1450 const struct ofproto_class *class;
1452 datapath_type = ofproto_normalize_type(datapath_type);
1453 class = ofproto_class_find__(datapath_type);
1458 return (class->port_open_type
1459 ? class->port_open_type(datapath_type, port_type)
1463 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1464 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1465 * the port's OpenFlow port number.
1467 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1468 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1469 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1470 * 'ofp_portp' is non-null). */
1472 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1473 uint16_t *ofp_portp)
1475 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1478 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1480 const char *netdev_name = netdev_get_name(netdev);
1482 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1483 update_port(ofproto, netdev_name);
1486 struct ofproto_port ofproto_port;
1488 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1490 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1491 ofproto_port_destroy(&ofproto_port);
1496 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1497 * initializes '*port' appropriately; on failure, returns a positive errno
1500 * The caller owns the data in 'ofproto_port' and must free it with
1501 * ofproto_port_destroy() when it is no longer needed. */
1503 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1504 struct ofproto_port *port)
1508 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1510 memset(port, 0, sizeof *port);
1515 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1516 * Returns 0 if successful, otherwise a positive errno. */
1518 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1520 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1521 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1522 struct simap_node *ofp_request_node;
1525 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1526 if (ofp_request_node) {
1527 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1530 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1531 if (!error && ofport) {
1532 /* 'name' is the netdev's name and update_port() is going to close the
1533 * netdev. Just in case update_port() refers to 'name' after it
1534 * destroys 'ofport', make a copy of it around the update_port()
1536 char *devname = xstrdup(name);
1537 update_port(ofproto, devname);
1543 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1544 * performs the 'n_actions' actions in 'actions'. The new flow will not
1547 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1548 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1549 * controllers; otherwise, it will be hidden.
1551 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1553 * This is a helper function for in-band control and fail-open. */
1555 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1556 unsigned int priority,
1557 const struct ofpact *ofpacts, size_t ofpacts_len)
1559 const struct rule *rule;
1561 rule = rule_from_cls_rule(classifier_find_match_exactly(
1562 &ofproto->tables[0].cls, match, priority));
1563 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1564 ofpacts, ofpacts_len)) {
1565 struct ofputil_flow_mod fm;
1567 memset(&fm, 0, sizeof fm);
1569 fm.priority = priority;
1570 fm.buffer_id = UINT32_MAX;
1571 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1572 fm.ofpacts_len = ofpacts_len;
1573 add_flow(ofproto, NULL, &fm, NULL);
1578 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1579 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1580 * operation cannot be initiated now but may be retried later.
1582 * This is a helper function for in-band control and fail-open. */
1584 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1586 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1589 /* Searches for a rule with matching criteria exactly equal to 'target' in
1590 * ofproto's table 0 and, if it finds one, deletes it.
1592 * This is a helper function for in-band control and fail-open. */
1594 ofproto_delete_flow(struct ofproto *ofproto,
1595 const struct match *target, unsigned int priority)
1599 rule = rule_from_cls_rule(classifier_find_match_exactly(
1600 &ofproto->tables[0].cls, target, priority));
1602 /* No such rule -> success. */
1604 } else if (rule->pending) {
1605 /* An operation on the rule is already pending -> failure.
1606 * Caller must retry later if it's important. */
1609 /* Initiate deletion -> success. */
1610 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1611 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1612 oftable_remove_rule(rule);
1613 ofproto->ofproto_class->rule_destruct(rule);
1614 ofopgroup_submit(group);
1620 /* Starts the process of deleting all of the flows from all of ofproto's flow
1621 * tables and then reintroducing the flows required by in-band control and
1622 * fail-open. The process will complete in a later call to ofproto_run(). */
1624 ofproto_flush_flows(struct ofproto *ofproto)
1626 COVERAGE_INC(ofproto_flush);
1627 ofproto->state = S_FLUSH;
1631 reinit_ports(struct ofproto *p)
1633 struct ofproto_port_dump dump;
1634 struct sset devnames;
1635 struct ofport *ofport;
1636 struct ofproto_port ofproto_port;
1637 const char *devname;
1639 COVERAGE_INC(ofproto_reinit_ports);
1641 sset_init(&devnames);
1642 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1643 sset_add(&devnames, netdev_get_name(ofport->netdev));
1645 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1646 sset_add(&devnames, ofproto_port.name);
1649 SSET_FOR_EACH (devname, &devnames) {
1650 update_port(p, devname);
1652 sset_destroy(&devnames);
1656 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1660 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1661 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1663 if (ofp_port >= ofproto->max_ports
1664 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1665 bool retry = ofproto->alloc_port_no ? true : false;
1667 /* Search for a free OpenFlow port number. We try not to
1668 * immediately reuse them to prevent problems due to old
1670 while (ofp_port >= ofproto->max_ports) {
1671 for (ofproto->alloc_port_no++;
1672 ofproto->alloc_port_no < ofproto->max_ports;
1673 ofproto->alloc_port_no++) {
1674 if (!bitmap_is_set(ofproto->ofp_port_ids,
1675 ofproto->alloc_port_no)) {
1676 ofp_port = ofproto->alloc_port_no;
1680 if (ofproto->alloc_port_no >= ofproto->max_ports) {
1682 ofproto->alloc_port_no = 0;
1691 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1696 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1698 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1701 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1702 * pointer if the netdev cannot be opened. On success, also fills in
1704 static struct netdev *
1705 ofport_open(struct ofproto *ofproto,
1706 struct ofproto_port *ofproto_port,
1707 struct ofputil_phy_port *pp)
1709 enum netdev_flags flags;
1710 struct netdev *netdev;
1713 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1715 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1716 "cannot be opened (%s)",
1718 ofproto_port->name, ofproto_port->ofp_port,
1719 ofproto_port->name, strerror(error));
1723 if (ofproto_port->ofp_port == OFPP_NONE) {
1724 if (!strcmp(ofproto->name, ofproto_port->name)) {
1725 ofproto_port->ofp_port = OFPP_LOCAL;
1727 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1728 ofproto_port->name);
1731 pp->port_no = ofproto_port->ofp_port;
1732 netdev_get_etheraddr(netdev, pp->hw_addr);
1733 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1734 netdev_get_flags(netdev, &flags);
1735 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1736 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1737 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1738 &pp->supported, &pp->peer);
1739 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1740 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1745 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1746 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1749 ofport_equal(const struct ofputil_phy_port *a,
1750 const struct ofputil_phy_port *b)
1752 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1753 && a->state == b->state
1754 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1755 && a->curr == b->curr
1756 && a->advertised == b->advertised
1757 && a->supported == b->supported
1758 && a->peer == b->peer
1759 && a->curr_speed == b->curr_speed
1760 && a->max_speed == b->max_speed);
1763 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1764 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1765 * one with the same name or port number). */
1767 ofport_install(struct ofproto *p,
1768 struct netdev *netdev, const struct ofputil_phy_port *pp)
1770 const char *netdev_name = netdev_get_name(netdev);
1771 struct ofport *ofport;
1774 /* Create ofport. */
1775 ofport = p->ofproto_class->port_alloc();
1780 ofport->ofproto = p;
1781 ofport->netdev = netdev;
1782 ofport->change_seq = netdev_change_seq(netdev);
1784 ofport->ofp_port = pp->port_no;
1786 /* Add port to 'p'. */
1787 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1788 shash_add(&p->port_by_name, netdev_name, ofport);
1790 update_mtu(p, ofport);
1792 /* Let the ofproto_class initialize its private data. */
1793 error = p->ofproto_class->port_construct(ofport);
1797 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1801 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1802 p->name, netdev_name, strerror(error));
1804 ofport_destroy__(ofport);
1806 netdev_close(netdev);
1810 /* Removes 'ofport' from 'p' and destroys it. */
1812 ofport_remove(struct ofport *ofport)
1814 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1816 ofport_destroy(ofport);
1819 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1822 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1824 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1826 ofport_remove(port);
1830 /* Updates 'port' with new 'pp' description.
1832 * Does not handle a name or port number change. The caller must implement
1833 * such a change as a delete followed by an add. */
1835 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1837 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1838 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1839 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1840 port->pp.state = pp->state;
1841 port->pp.curr = pp->curr;
1842 port->pp.advertised = pp->advertised;
1843 port->pp.supported = pp->supported;
1844 port->pp.peer = pp->peer;
1845 port->pp.curr_speed = pp->curr_speed;
1846 port->pp.max_speed = pp->max_speed;
1848 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1851 /* Update OpenFlow 'state' in 'port' and notify controller. */
1853 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1855 if (port->pp.state != state) {
1856 port->pp.state = state;
1857 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1863 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1865 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1867 if (port->ofproto->ofproto_class->set_realdev) {
1868 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1870 if (port->ofproto->ofproto_class->set_stp_port) {
1871 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1873 if (port->ofproto->ofproto_class->set_cfm) {
1874 port->ofproto->ofproto_class->set_cfm(port, NULL);
1876 if (port->ofproto->ofproto_class->bundle_remove) {
1877 port->ofproto->ofproto_class->bundle_remove(port);
1883 ofport_destroy__(struct ofport *port)
1885 struct ofproto *ofproto = port->ofproto;
1886 const char *name = netdev_get_name(port->netdev);
1888 hmap_remove(&ofproto->ports, &port->hmap_node);
1889 shash_delete(&ofproto->port_by_name,
1890 shash_find(&ofproto->port_by_name, name));
1892 netdev_close(port->netdev);
1893 ofproto->ofproto_class->port_dealloc(port);
1897 ofport_destroy(struct ofport *port)
1900 dealloc_ofp_port(port->ofproto, port->ofp_port);
1901 port->ofproto->ofproto_class->port_destruct(port);
1902 ofport_destroy__(port);
1907 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1909 struct ofport *port;
1911 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1912 hash_int(ofp_port, 0), &ofproto->ports) {
1913 if (port->ofp_port == ofp_port) {
1921 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1923 struct ofproto *ofproto = port->ofproto;
1926 if (ofproto->ofproto_class->port_get_stats) {
1927 error = ofproto->ofproto_class->port_get_stats(port, stats);
1936 update_port(struct ofproto *ofproto, const char *name)
1938 struct ofproto_port ofproto_port;
1939 struct ofputil_phy_port pp;
1940 struct netdev *netdev;
1941 struct ofport *port;
1943 COVERAGE_INC(ofproto_update_port);
1945 /* Fetch 'name''s location and properties from the datapath. */
1946 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1947 ? ofport_open(ofproto, &ofproto_port, &pp)
1950 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1951 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1952 struct netdev *old_netdev = port->netdev;
1954 /* 'name' hasn't changed location. Any properties changed? */
1955 if (!ofport_equal(&port->pp, &pp)) {
1956 ofport_modified(port, &pp);
1959 update_mtu(ofproto, port);
1961 /* Install the newly opened netdev in case it has changed.
1962 * Don't close the old netdev yet in case port_modified has to
1963 * remove a retained reference to it.*/
1964 port->netdev = netdev;
1965 port->change_seq = netdev_change_seq(netdev);
1967 if (port->ofproto->ofproto_class->port_modified) {
1968 port->ofproto->ofproto_class->port_modified(port);
1971 netdev_close(old_netdev);
1973 /* If 'port' is nonnull then its name differs from 'name' and thus
1974 * we should delete it. If we think there's a port named 'name'
1975 * then its port number must be wrong now so delete it too. */
1977 ofport_remove(port);
1979 ofport_remove_with_name(ofproto, name);
1980 ofport_install(ofproto, netdev, &pp);
1983 /* Any port named 'name' is gone now. */
1984 ofport_remove_with_name(ofproto, name);
1986 ofproto_port_destroy(&ofproto_port);
1990 init_ports(struct ofproto *p)
1992 struct ofproto_port_dump dump;
1993 struct ofproto_port ofproto_port;
1994 struct shash_node *node, *next;
1996 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1997 const char *name = ofproto_port.name;
1999 if (shash_find(&p->port_by_name, name)) {
2000 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2003 struct ofputil_phy_port pp;
2004 struct netdev *netdev;
2006 /* Check if an OpenFlow port number had been requested. */
2007 node = shash_find(&init_ofp_ports, name);
2009 const struct iface_hint *iface_hint = node->data;
2010 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
2013 netdev = ofport_open(p, &ofproto_port, &pp);
2015 ofport_install(p, netdev, &pp);
2020 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2021 struct iface_hint *iface_hint = node->data;
2023 if (!strcmp(iface_hint->br_name, p->name)) {
2024 free(iface_hint->br_name);
2025 free(iface_hint->br_type);
2027 shash_delete(&init_ofp_ports, node);
2034 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2035 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2037 find_min_mtu(struct ofproto *p)
2039 struct ofport *ofport;
2042 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2043 struct netdev *netdev = ofport->netdev;
2046 /* Skip any internal ports, since that's what we're trying to
2048 if (!strcmp(netdev_get_type(netdev), "internal")) {
2052 if (netdev_get_mtu(netdev, &dev_mtu)) {
2055 if (!mtu || dev_mtu < mtu) {
2060 return mtu ? mtu: ETH_PAYLOAD_MAX;
2063 /* Update MTU of all datapath devices on 'p' to the minimum of the
2064 * non-datapath ports in event of 'port' added or changed. */
2066 update_mtu(struct ofproto *p, struct ofport *port)
2068 struct ofport *ofport;
2069 struct netdev *netdev = port->netdev;
2070 int dev_mtu, old_min;
2072 if (netdev_get_mtu(netdev, &dev_mtu)) {
2076 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2077 if (dev_mtu > p->min_mtu) {
2078 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2079 dev_mtu = p->min_mtu;
2082 port->mtu = dev_mtu;
2086 /* For non-internal port find new min mtu. */
2087 old_min = p->min_mtu;
2088 port->mtu = dev_mtu;
2089 p->min_mtu = find_min_mtu(p);
2090 if (p->min_mtu == old_min) {
2094 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2095 struct netdev *netdev = ofport->netdev;
2097 if (!strcmp(netdev_get_type(netdev), "internal")) {
2098 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2099 ofport->mtu = p->min_mtu;
2106 ofproto_rule_destroy__(struct rule *rule)
2109 cls_rule_destroy(&rule->cr);
2110 free(rule->ofpacts);
2111 rule->ofproto->ofproto_class->rule_dealloc(rule);
2115 /* This function allows an ofproto implementation to destroy any rules that
2116 * remain when its ->destruct() function is called. The caller must have
2117 * already uninitialized any derived members of 'rule' (step 5 described in the
2118 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2119 * This function implements steps 6 and 7.
2121 * This function should only be called from an ofproto implementation's
2122 * ->destruct() function. It is not suitable elsewhere. */
2124 ofproto_rule_destroy(struct rule *rule)
2126 assert(!rule->pending);
2127 oftable_remove_rule(rule);
2128 ofproto_rule_destroy__(rule);
2131 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2132 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2134 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2136 return (port == OFPP_ANY
2137 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2140 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2141 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2143 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2145 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2150 case OFOPERATION_ADD:
2151 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2153 case OFOPERATION_DELETE:
2156 case OFOPERATION_MODIFY:
2157 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2163 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2164 * statistics appropriately. 'packet' must have at least sizeof(struct
2165 * ofp10_packet_in) bytes of headroom.
2167 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2168 * with statistics for 'packet' either way.
2170 * Takes ownership of 'packet'. */
2172 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2176 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2178 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2179 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2182 /* Returns true if 'rule' should be hidden from the controller.
2184 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2185 * (e.g. by in-band control) and are intentionally hidden from the
2188 ofproto_rule_is_hidden(const struct rule *rule)
2190 return rule->cr.priority > UINT16_MAX;
2193 static enum oftable_flags
2194 rule_get_flags(const struct rule *rule)
2196 return rule->ofproto->tables[rule->table_id].flags;
2200 rule_is_modifiable(const struct rule *rule)
2202 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2206 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2208 ofconn_send_reply(ofconn, make_echo_reply(oh));
2213 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2215 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2216 struct ofputil_switch_features features;
2217 struct ofport *port;
2223 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2225 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2227 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2228 * if present, are always at the end.) */
2229 n_tables = ofproto->n_tables;
2230 for (i = 0; i < ofproto->n_tables; i++) {
2231 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2237 features.datapath_id = ofproto->datapath_id;
2238 features.n_buffers = pktbuf_capacity();
2239 features.n_tables = n_tables;
2240 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2241 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2243 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2245 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2246 features.auxiliary_id = 0;
2247 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2249 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2250 ofputil_put_switch_features_port(&port->pp, b);
2253 ofconn_send_reply(ofconn, b);
2258 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2260 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2261 struct ofp_switch_config *osc;
2262 enum ofp_config_flags flags;
2266 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2267 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2268 flags = ofproto->frag_handling;
2269 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2270 if (oh->version < OFP13_VERSION
2271 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2272 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2274 osc->flags = htons(flags);
2275 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2276 ofconn_send_reply(ofconn, buf);
2282 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2284 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2285 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2286 uint16_t flags = ntohs(osc->flags);
2288 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2289 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2290 enum ofp_config_flags cur = ofproto->frag_handling;
2291 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2293 assert((cur & OFPC_FRAG_MASK) == cur);
2295 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2296 ofproto->frag_handling = next;
2298 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2300 ofputil_frag_handling_to_string(next));
2304 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2305 ofconn_set_invalid_ttl_to_controller(ofconn,
2306 (oh->version < OFP13_VERSION
2307 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2309 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2314 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2315 * error message code for the caller to propagate upward. Otherwise, returns
2318 * The log message mentions 'msg_type'. */
2320 reject_slave_controller(struct ofconn *ofconn)
2322 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2323 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2324 return OFPERR_OFPBRC_EPERM;
2331 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2333 struct ofproto *p = ofconn_get_ofproto(ofconn);
2334 struct ofputil_packet_out po;
2335 struct ofpbuf *payload;
2336 uint64_t ofpacts_stub[1024 / 8];
2337 struct ofpbuf ofpacts;
2341 COVERAGE_INC(ofproto_packet_out);
2343 error = reject_slave_controller(ofconn);
2348 /* Decode message. */
2349 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2350 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2352 goto exit_free_ofpacts;
2354 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2355 error = OFPERR_OFPBRC_BAD_PORT;
2356 goto exit_free_ofpacts;
2361 if (po.buffer_id != UINT32_MAX) {
2362 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2363 if (error || !payload) {
2364 goto exit_free_ofpacts;
2367 payload = xmalloc(sizeof *payload);
2368 ofpbuf_use_const(payload, po.packet, po.packet_len);
2371 /* Verify actions against packet, then send packet if successful. */
2372 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2373 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2375 error = p->ofproto_class->packet_out(p, payload, &flow,
2376 po.ofpacts, po.ofpacts_len);
2378 ofpbuf_delete(payload);
2381 ofpbuf_uninit(&ofpacts);
2387 update_port_config(struct ofport *port,
2388 enum ofputil_port_config config,
2389 enum ofputil_port_config mask)
2391 enum ofputil_port_config old_config = port->pp.config;
2392 enum ofputil_port_config toggle;
2394 toggle = (config ^ port->pp.config) & mask;
2395 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2396 if (config & OFPUTIL_PC_PORT_DOWN) {
2397 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2399 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2401 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2404 port->pp.config ^= toggle;
2405 if (port->pp.config != old_config) {
2406 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2411 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2413 struct ofproto *p = ofconn_get_ofproto(ofconn);
2414 struct ofputil_port_mod pm;
2415 struct ofport *port;
2418 error = reject_slave_controller(ofconn);
2423 error = ofputil_decode_port_mod(oh, &pm);
2428 port = ofproto_get_port(p, pm.port_no);
2430 return OFPERR_OFPPMFC_BAD_PORT;
2431 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2432 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2434 update_port_config(port, pm.config, pm.mask);
2436 netdev_set_advertisements(port->netdev, pm.advertise);
2443 handle_desc_stats_request(struct ofconn *ofconn,
2444 const struct ofp_header *request)
2446 struct ofproto *p = ofconn_get_ofproto(ofconn);
2447 struct ofp_desc_stats *ods;
2450 msg = ofpraw_alloc_stats_reply(request, 0);
2451 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2452 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2453 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2454 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2455 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2456 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2457 ofconn_send_reply(ofconn, msg);
2463 handle_table_stats_request(struct ofconn *ofconn,
2464 const struct ofp_header *request)
2466 struct ofproto *p = ofconn_get_ofproto(ofconn);
2467 struct ofp12_table_stats *ots;
2472 /* Set up default values.
2474 * ofp12_table_stats is used as a generic structure as
2475 * it is able to hold all the fields for ofp10_table_stats
2476 * and ofp11_table_stats (and of course itself).
2478 ots = xcalloc(p->n_tables, sizeof *ots);
2479 for (i = 0; i < p->n_tables; i++) {
2480 ots[i].table_id = i;
2481 sprintf(ots[i].name, "table%zu", i);
2482 ots[i].match = htonll(OFPXMT12_MASK);
2483 ots[i].wildcards = htonll(OFPXMT12_MASK);
2484 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2485 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2486 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2487 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2488 ots[i].metadata_match = htonll(UINT64_MAX);
2489 ots[i].metadata_write = htonll(UINT64_MAX);
2490 ots[i].instructions = htonl(OFPIT11_ALL);
2491 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2492 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2493 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2496 p->ofproto_class->get_tables(p, ots);
2498 /* Post-process the tables, dropping hidden tables. */
2499 n_tables = p->n_tables;
2500 for (i = 0; i < p->n_tables; i++) {
2501 const struct oftable *table = &p->tables[i];
2503 if (table->flags & OFTABLE_HIDDEN) {
2509 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2512 if (table->max_flows < ntohl(ots[i].max_entries)) {
2513 ots[i].max_entries = htonl(table->max_flows);
2517 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2518 ofconn_send_reply(ofconn, msg);
2526 append_port_stat(struct ofport *port, struct list *replies)
2528 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2530 /* Intentionally ignore return value, since errors will set
2531 * 'stats' to all-1s, which is correct for OpenFlow, and
2532 * netdev_get_stats() will log errors. */
2533 ofproto_port_get_stats(port, &ops.stats);
2535 ofputil_append_port_stat(replies, &ops);
2539 handle_port_stats_request(struct ofconn *ofconn,
2540 const struct ofp_header *request)
2542 struct ofproto *p = ofconn_get_ofproto(ofconn);
2543 struct ofport *port;
2544 struct list replies;
2548 error = ofputil_decode_port_stats_request(request, &port_no);
2553 ofpmp_init(&replies, request);
2554 if (port_no != OFPP_ANY) {
2555 port = ofproto_get_port(p, port_no);
2557 append_port_stat(port, &replies);
2560 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2561 append_port_stat(port, &replies);
2565 ofconn_send_replies(ofconn, &replies);
2570 handle_port_desc_stats_request(struct ofconn *ofconn,
2571 const struct ofp_header *request)
2573 struct ofproto *p = ofconn_get_ofproto(ofconn);
2574 enum ofp_version version;
2575 struct ofport *port;
2576 struct list replies;
2578 ofpmp_init(&replies, request);
2580 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2581 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2582 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2585 ofconn_send_replies(ofconn, &replies);
2590 calc_flow_duration__(long long int start, long long int now,
2591 uint32_t *sec, uint32_t *nsec)
2593 long long int msecs = now - start;
2594 *sec = msecs / 1000;
2595 *nsec = (msecs % 1000) * (1000 * 1000);
2598 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2599 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2601 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2603 return (table_id == 0xff || table_id < ofproto->n_tables
2605 : OFPERR_OFPBRC_BAD_TABLE_ID);
2609 static struct oftable *
2610 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2612 struct oftable *table;
2614 for (table = &ofproto->tables[table_id];
2615 table < &ofproto->tables[ofproto->n_tables];
2617 if (!(table->flags & OFTABLE_HIDDEN)) {
2625 static struct oftable *
2626 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2628 if (table_id == 0xff) {
2629 return next_visible_table(ofproto, 0);
2630 } else if (table_id < ofproto->n_tables) {
2631 return &ofproto->tables[table_id];
2637 static struct oftable *
2638 next_matching_table(const struct ofproto *ofproto,
2639 const struct oftable *table, uint8_t table_id)
2641 return (table_id == 0xff
2642 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2646 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2648 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2649 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2651 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2652 * only once, for that table. (This can be used to access tables marked
2655 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2656 * entered at all. (Perhaps you should have validated TABLE_ID with
2657 * check_table_id().)
2659 * All parameters are evaluated multiple times.
2661 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2662 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2664 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2666 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2667 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2668 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2671 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2672 * to 'out_port' are included.
2674 * Hidden rules are always omitted.
2676 * Returns 0 on success, otherwise an OpenFlow error code. */
2678 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2679 const struct match *match,
2680 ovs_be64 cookie, ovs_be64 cookie_mask,
2681 uint16_t out_port, struct list *rules)
2683 struct oftable *table;
2687 error = check_table_id(ofproto, table_id);
2693 cls_rule_init(&cr, match, 0);
2694 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2695 struct cls_cursor cursor;
2698 cls_cursor_init(&cursor, &table->cls, &cr);
2699 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2700 if (rule->pending) {
2701 error = OFPROTO_POSTPONE;
2704 if (!ofproto_rule_is_hidden(rule)
2705 && ofproto_rule_has_out_port(rule, out_port)
2706 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2707 list_push_back(rules, &rule->ofproto_node);
2713 cls_rule_destroy(&cr);
2717 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2718 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2719 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2722 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2723 * to 'out_port' are included.
2725 * Hidden rules are always omitted.
2727 * Returns 0 on success, otherwise an OpenFlow error code. */
2729 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2730 const struct match *match, unsigned int priority,
2731 ovs_be64 cookie, ovs_be64 cookie_mask,
2732 uint16_t out_port, struct list *rules)
2734 struct oftable *table;
2738 error = check_table_id(ofproto, table_id);
2744 cls_rule_init(&cr, match, priority);
2745 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2748 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2751 if (rule->pending) {
2752 error = OFPROTO_POSTPONE;
2755 if (!ofproto_rule_is_hidden(rule)
2756 && ofproto_rule_has_out_port(rule, out_port)
2757 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2758 list_push_back(rules, &rule->ofproto_node);
2764 cls_rule_destroy(&cr);
2768 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2769 * forced into the range of a uint16_t. */
2771 age_secs(long long int age_ms)
2773 return (age_ms < 0 ? 0
2774 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2775 : (unsigned int) age_ms / 1000);
2779 handle_flow_stats_request(struct ofconn *ofconn,
2780 const struct ofp_header *request)
2782 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2783 struct ofputil_flow_stats_request fsr;
2784 struct list replies;
2789 error = ofputil_decode_flow_stats_request(&fsr, request);
2794 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2795 fsr.cookie, fsr.cookie_mask,
2796 fsr.out_port, &rules);
2801 ofpmp_init(&replies, request);
2802 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2803 long long int now = time_msec();
2804 struct ofputil_flow_stats fs;
2806 minimatch_expand(&rule->cr.match, &fs.match);
2807 fs.priority = rule->cr.priority;
2808 fs.cookie = rule->flow_cookie;
2809 fs.table_id = rule->table_id;
2810 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2812 fs.idle_timeout = rule->idle_timeout;
2813 fs.hard_timeout = rule->hard_timeout;
2814 fs.idle_age = age_secs(now - rule->used);
2815 fs.hard_age = age_secs(now - rule->modified);
2816 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2818 fs.ofpacts = rule->ofpacts;
2819 fs.ofpacts_len = rule->ofpacts_len;
2821 if (rule->send_flow_removed) {
2822 fs.flags |= OFPFF_SEND_FLOW_REM;
2823 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2824 and OFPFF13_NO_BYT_COUNTS */
2826 ofputil_append_flow_stats_reply(&fs, &replies);
2828 ofconn_send_replies(ofconn, &replies);
2834 flow_stats_ds(struct rule *rule, struct ds *results)
2836 uint64_t packet_count, byte_count;
2838 rule->ofproto->ofproto_class->rule_get_stats(rule,
2839 &packet_count, &byte_count);
2841 if (rule->table_id != 0) {
2842 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2844 ds_put_format(results, "duration=%llds, ",
2845 (time_msec() - rule->created) / 1000);
2846 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2847 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2848 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2849 cls_rule_format(&rule->cr, results);
2850 ds_put_char(results, ',');
2851 if (rule->ofpacts_len > 0) {
2852 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2854 ds_put_cstr(results, "drop");
2856 ds_put_cstr(results, "\n");
2859 /* Adds a pretty-printed description of all flows to 'results', including
2860 * hidden flows (e.g., set up by in-band control). */
2862 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2864 struct oftable *table;
2866 OFPROTO_FOR_EACH_TABLE (table, p) {
2867 struct cls_cursor cursor;
2870 cls_cursor_init(&cursor, &table->cls, NULL);
2871 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2872 flow_stats_ds(rule, results);
2877 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2878 * '*engine_type' and '*engine_id', respectively. */
2880 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2881 uint8_t *engine_type, uint8_t *engine_id)
2883 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2886 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2887 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2888 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2889 * and -1 if CFM is not enabled on 'ofp_port'. */
2891 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2893 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2894 return (ofport && ofproto->ofproto_class->get_cfm_fault
2895 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2899 /* Checks the operational status reported by the remote CFM endpoint of
2900 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2901 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2904 ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
2906 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2907 return (ofport && ofproto->ofproto_class->get_cfm_opup
2908 ? ofproto->ofproto_class->get_cfm_opup(ofport)
2912 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2913 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2914 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2915 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2917 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2918 uint16_t ofp_port, const uint64_t **rmps,
2921 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2925 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2926 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2931 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2932 * integer value between 0 and 100 to indicate the health of the port as a
2933 * percentage which is the average of cfm health of all the remote_mpids or
2934 * returns -1 if CFM is not enabled on 'ofport'. */
2936 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2938 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2939 return (ofport && ofproto->ofproto_class->get_cfm_health
2940 ? ofproto->ofproto_class->get_cfm_health(ofport)
2945 handle_aggregate_stats_request(struct ofconn *ofconn,
2946 const struct ofp_header *oh)
2948 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2949 struct ofputil_flow_stats_request request;
2950 struct ofputil_aggregate_stats stats;
2951 bool unknown_packets, unknown_bytes;
2952 struct ofpbuf *reply;
2957 error = ofputil_decode_flow_stats_request(&request, oh);
2962 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2963 request.cookie, request.cookie_mask,
2964 request.out_port, &rules);
2969 memset(&stats, 0, sizeof stats);
2970 unknown_packets = unknown_bytes = false;
2971 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2972 uint64_t packet_count;
2973 uint64_t byte_count;
2975 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2978 if (packet_count == UINT64_MAX) {
2979 unknown_packets = true;
2981 stats.packet_count += packet_count;
2984 if (byte_count == UINT64_MAX) {
2985 unknown_bytes = true;
2987 stats.byte_count += byte_count;
2992 if (unknown_packets) {
2993 stats.packet_count = UINT64_MAX;
2995 if (unknown_bytes) {
2996 stats.byte_count = UINT64_MAX;
2999 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3000 ofconn_send_reply(ofconn, reply);
3005 struct queue_stats_cbdata {
3006 struct ofport *ofport;
3007 struct list replies;
3011 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3012 const struct netdev_queue_stats *stats)
3015 struct ofputil_queue_stats oqs = {
3016 .port_no = cbdata->ofport->pp.port_no,
3017 .queue_id = queue_id,
3020 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3024 handle_queue_stats_dump_cb(uint32_t queue_id,
3025 struct netdev_queue_stats *stats,
3028 struct queue_stats_cbdata *cbdata = cbdata_;
3030 put_queue_stats(cbdata, queue_id, stats);
3034 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3035 struct queue_stats_cbdata *cbdata)
3037 cbdata->ofport = port;
3038 if (queue_id == OFPQ_ALL) {
3039 netdev_dump_queue_stats(port->netdev,
3040 handle_queue_stats_dump_cb, cbdata);
3042 struct netdev_queue_stats stats;
3044 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3045 put_queue_stats(cbdata, queue_id, &stats);
3047 return OFPERR_OFPQOFC_BAD_QUEUE;
3054 handle_queue_stats_request(struct ofconn *ofconn,
3055 const struct ofp_header *rq)
3057 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3058 struct queue_stats_cbdata cbdata;
3059 struct ofport *port;
3061 struct ofputil_queue_stats_request oqsr;
3063 COVERAGE_INC(ofproto_queue_req);
3065 ofpmp_init(&cbdata.replies, rq);
3067 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3072 if (oqsr.port_no == OFPP_ANY) {
3073 error = OFPERR_OFPQOFC_BAD_QUEUE;
3074 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3075 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3080 port = ofproto_get_port(ofproto, oqsr.port_no);
3082 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3083 : OFPERR_OFPQOFC_BAD_PORT);
3086 ofconn_send_replies(ofconn, &cbdata.replies);
3088 ofpbuf_list_delete(&cbdata.replies);
3095 is_flow_deletion_pending(const struct ofproto *ofproto,
3096 const struct cls_rule *cls_rule,
3099 if (!hmap_is_empty(&ofproto->deletions)) {
3100 struct ofoperation *op;
3102 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3103 cls_rule_hash(cls_rule, table_id),
3104 &ofproto->deletions) {
3105 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3114 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3115 * in which no matching flow already exists in the flow table.
3117 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3118 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3119 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3120 * initiated now but may be retried later.
3122 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3123 * ownership remains with the caller.
3125 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3128 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3129 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3131 struct oftable *table;
3132 struct ofopgroup *group;
3133 struct rule *victim;
3138 error = check_table_id(ofproto, fm->table_id);
3144 if (fm->table_id == 0xff) {
3146 if (ofproto->ofproto_class->rule_choose_table) {
3147 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3153 assert(table_id < ofproto->n_tables);
3154 table = &ofproto->tables[table_id];
3156 table = &ofproto->tables[0];
3158 } else if (fm->table_id < ofproto->n_tables) {
3159 table = &ofproto->tables[fm->table_id];
3161 return OFPERR_OFPBRC_BAD_TABLE_ID;
3164 if (table->flags & OFTABLE_READONLY) {
3165 return OFPERR_OFPBRC_EPERM;
3168 /* Allocate new rule and initialize classifier rule. */
3169 rule = ofproto->ofproto_class->rule_alloc();
3171 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3172 ofproto->name, strerror(error));
3175 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3177 /* Serialize against pending deletion. */
3178 if (is_flow_deletion_pending(ofproto, &cr, table - ofproto->tables)) {
3179 cls_rule_destroy(&rule->cr);
3180 ofproto->ofproto_class->rule_dealloc(rule);
3181 return OFPROTO_POSTPONE;
3184 /* Check for overlap, if requested. */
3185 if (fm->flags & OFPFF_CHECK_OVERLAP
3186 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3187 cls_rule_destroy(&rule->cr);
3188 ofproto->ofproto_class->rule_dealloc(rule);
3189 return OFPERR_OFPFMFC_OVERLAP;
3192 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3194 rule->ofproto = ofproto;
3195 rule->pending = NULL;
3196 rule->flow_cookie = fm->new_cookie;
3197 rule->created = rule->modified = rule->used = time_msec();
3198 rule->idle_timeout = fm->idle_timeout;
3199 rule->hard_timeout = fm->hard_timeout;
3200 rule->table_id = table - ofproto->tables;
3201 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3202 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3203 and OFPFF13_NO_BYT_COUNTS */
3204 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3205 rule->ofpacts_len = fm->ofpacts_len;
3206 rule->evictable = true;
3207 rule->eviction_group = NULL;
3208 rule->monitor_flags = 0;
3209 rule->add_seqno = 0;
3210 rule->modify_seqno = 0;
3212 /* Insert new rule. */
3213 victim = oftable_replace_rule(rule);
3214 if (victim && !rule_is_modifiable(victim)) {
3215 error = OFPERR_OFPBRC_EPERM;
3216 } else if (victim && victim->pending) {
3217 error = OFPROTO_POSTPONE;
3219 struct ofoperation *op;
3222 if (classifier_count(&table->cls) > table->max_flows) {
3225 was_evictable = rule->evictable;
3226 rule->evictable = false;
3227 evict = choose_rule_to_evict(table);
3228 rule->evictable = was_evictable;
3231 error = OFPERR_OFPFMFC_TABLE_FULL;
3233 } else if (evict->pending) {
3234 error = OFPROTO_POSTPONE;
3241 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3242 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3243 op->victim = victim;
3245 error = ofproto->ofproto_class->rule_construct(rule);
3247 op->group->n_running--;
3248 ofoperation_destroy(rule->pending);
3250 delete_flow__(evict, group);
3252 ofopgroup_submit(group);
3256 /* Back out if an error occurred. */
3258 oftable_substitute_rule(rule, victim);
3259 ofproto_rule_destroy__(rule);
3264 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3266 /* Modifies the rules listed in 'rules', changing their actions to match those
3269 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3272 * Returns 0 on success, otherwise an OpenFlow error code. */
3274 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3275 const struct ofputil_flow_mod *fm,
3276 const struct ofp_header *request, struct list *rules)
3278 struct ofopgroup *group;
3282 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3283 error = OFPERR_OFPBRC_EPERM;
3284 LIST_FOR_EACH (rule, ofproto_node, rules) {
3285 struct ofoperation *op;
3286 bool actions_changed;
3287 ovs_be64 new_cookie;
3289 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3291 if (rule_is_modifiable(rule)) {
3292 /* At least one rule is modifiable, don't report EPERM error. */
3298 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3299 rule->ofpacts, rule->ofpacts_len);
3300 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3302 : rule->flow_cookie);
3303 if (!actions_changed && new_cookie == rule->flow_cookie) {
3304 /* No change at all. */
3308 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3309 rule->flow_cookie = new_cookie;
3310 if (actions_changed) {
3311 op->ofpacts = rule->ofpacts;
3312 op->ofpacts_len = rule->ofpacts_len;
3313 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3314 rule->ofpacts_len = fm->ofpacts_len;
3315 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3317 ofoperation_complete(op, 0);
3320 ofopgroup_submit(group);
3326 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3327 const struct ofputil_flow_mod *fm,
3328 const struct ofp_header *request)
3330 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3333 return add_flow(ofproto, ofconn, fm, request);
3336 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3339 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3342 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3343 const struct ofputil_flow_mod *fm,
3344 const struct ofp_header *request)
3349 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3350 fm->cookie, fm->cookie_mask,
3354 } else if (list_is_empty(&rules)) {
3355 return modify_flows_add(ofproto, ofconn, fm, request);
3357 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3361 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3364 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3367 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3368 const struct ofputil_flow_mod *fm,
3369 const struct ofp_header *request)
3374 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3375 fm->priority, fm->cookie, fm->cookie_mask,
3380 } else if (list_is_empty(&rules)) {
3381 return modify_flows_add(ofproto, ofconn, fm, request);
3383 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3384 fm, request, &rules)
3389 /* OFPFC_DELETE implementation. */
3392 delete_flow__(struct rule *rule, struct ofopgroup *group)
3394 struct ofproto *ofproto = rule->ofproto;
3396 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3398 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3399 oftable_remove_rule(rule);
3400 ofproto->ofproto_class->rule_destruct(rule);
3403 /* Deletes the rules listed in 'rules'.
3405 * Returns 0 on success, otherwise an OpenFlow error code. */
3407 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3408 const struct ofp_header *request, struct list *rules)
3410 struct rule *rule, *next;
3411 struct ofopgroup *group;
3413 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3414 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3415 delete_flow__(rule, group);
3417 ofopgroup_submit(group);
3422 /* Implements OFPFC_DELETE. */
3424 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3425 const struct ofputil_flow_mod *fm,
3426 const struct ofp_header *request)
3431 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3432 fm->cookie, fm->cookie_mask,
3433 fm->out_port, &rules);
3434 return (error ? error
3435 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3440 /* Implements OFPFC_DELETE_STRICT. */
3442 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3443 const struct ofputil_flow_mod *fm,
3444 const struct ofp_header *request)
3449 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3450 fm->priority, fm->cookie, fm->cookie_mask,
3451 fm->out_port, &rules);
3452 return (error ? error
3453 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3459 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3461 struct ofputil_flow_removed fr;
3463 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3467 minimatch_expand(&rule->cr.match, &fr.match);
3468 fr.priority = rule->cr.priority;
3469 fr.cookie = rule->flow_cookie;
3471 fr.table_id = rule->table_id;
3472 calc_flow_duration__(rule->created, time_msec(),
3473 &fr.duration_sec, &fr.duration_nsec);
3474 fr.idle_timeout = rule->idle_timeout;
3475 fr.hard_timeout = rule->hard_timeout;
3476 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3479 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3483 ofproto_rule_update_used(struct rule *rule, long long int used)
3485 if (used > rule->used) {
3486 struct eviction_group *evg = rule->eviction_group;
3490 heap_change(&evg->rules, &rule->evg_node,
3491 rule_eviction_priority(rule));
3496 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3497 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3500 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3503 * ofproto implementation ->run() functions should use this function to expire
3504 * OpenFlow flows. */
3506 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3508 struct ofproto *ofproto = rule->ofproto;
3509 struct ofopgroup *group;
3511 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3513 ofproto_rule_send_removed(rule, reason);
3515 group = ofopgroup_create_unattached(ofproto);
3516 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3517 oftable_remove_rule(rule);
3518 ofproto->ofproto_class->rule_destruct(rule);
3519 ofopgroup_submit(group);
3523 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3525 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3526 struct ofputil_flow_mod fm;
3527 uint64_t ofpacts_stub[1024 / 8];
3528 struct ofpbuf ofpacts;
3532 error = reject_slave_controller(ofconn);
3537 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3538 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3541 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3542 &fm.match.flow, ofproto->max_ports);
3545 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3548 goto exit_free_ofpacts;
3551 /* Record the operation for logging a summary report. */
3552 switch (fm.command) {
3558 case OFPFC_MODIFY_STRICT:
3559 ofproto->n_modify++;
3563 case OFPFC_DELETE_STRICT:
3564 ofproto->n_delete++;
3569 if (ofproto->next_op_report == LLONG_MAX) {
3570 ofproto->first_op = now;
3571 ofproto->next_op_report = MAX(now + 10 * 1000,
3572 ofproto->op_backoff);
3573 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3575 ofproto->last_op = now;
3578 ofpbuf_uninit(&ofpacts);
3584 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3585 const struct ofputil_flow_mod *fm,
3586 const struct ofp_header *oh)
3588 if (ofproto->n_pending >= 50) {
3589 assert(!list_is_empty(&ofproto->pending));
3590 return OFPROTO_POSTPONE;
3593 switch (fm->command) {
3595 return add_flow(ofproto, ofconn, fm, oh);
3598 return modify_flows_loose(ofproto, ofconn, fm, oh);
3600 case OFPFC_MODIFY_STRICT:
3601 return modify_flow_strict(ofproto, ofconn, fm, oh);
3604 return delete_flows_loose(ofproto, ofconn, fm, oh);
3606 case OFPFC_DELETE_STRICT:
3607 return delete_flow_strict(ofproto, ofconn, fm, oh);
3610 if (fm->command > 0xff) {
3611 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3612 "flow_mod_table_id extension is not enabled",
3615 return OFPERR_OFPFMFC_BAD_COMMAND;
3620 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3622 const struct nx_role_request *nrr = ofpmsg_body(oh);
3623 struct nx_role_request *reply;
3627 role = ntohl(nrr->role);
3628 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3629 && role != NX_ROLE_SLAVE) {
3630 return OFPERR_OFPRRFC_BAD_ROLE;
3633 if (ofconn_get_role(ofconn) != role
3634 && ofconn_has_pending_opgroups(ofconn)) {
3635 return OFPROTO_POSTPONE;
3638 ofconn_set_role(ofconn, role);
3640 buf = ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY, oh, 0);
3641 reply = ofpbuf_put_zeros(buf, sizeof *reply);
3642 reply->role = htonl(role);
3643 ofconn_send_reply(ofconn, buf);
3649 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3650 const struct ofp_header *oh)
3652 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3653 enum ofputil_protocol cur, next;
3655 cur = ofconn_get_protocol(ofconn);
3656 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3657 ofconn_set_protocol(ofconn, next);
3663 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3665 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3666 enum ofputil_protocol cur, next;
3667 enum ofputil_protocol next_base;
3669 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3671 return OFPERR_OFPBRC_EPERM;
3674 cur = ofconn_get_protocol(ofconn);
3675 next = ofputil_protocol_set_base(cur, next_base);
3676 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3677 /* Avoid sending async messages in surprising protocol. */
3678 return OFPROTO_POSTPONE;
3681 ofconn_set_protocol(ofconn, next);
3686 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3687 const struct ofp_header *oh)
3689 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3692 format = ntohl(msg->format);
3693 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3694 return OFPERR_OFPBRC_EPERM;
3697 if (format != ofconn_get_packet_in_format(ofconn)
3698 && ofconn_has_pending_opgroups(ofconn)) {
3699 /* Avoid sending async message in surprsing packet in format. */
3700 return OFPROTO_POSTPONE;
3703 ofconn_set_packet_in_format(ofconn, format);
3708 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3710 const struct nx_async_config *msg = ofpmsg_body(oh);
3711 uint32_t master[OAM_N_TYPES];
3712 uint32_t slave[OAM_N_TYPES];
3714 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3715 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3716 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3718 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3719 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3720 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3722 ofconn_set_async_config(ofconn, master, slave);
3723 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3724 !ofconn_get_miss_send_len(ofconn)) {
3725 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3732 handle_nxt_set_controller_id(struct ofconn *ofconn,
3733 const struct ofp_header *oh)
3735 const struct nx_controller_id *nci = ofpmsg_body(oh);
3737 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3738 return OFPERR_NXBRC_MUST_BE_ZERO;
3741 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3746 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3750 if (ofconn_has_pending_opgroups(ofconn)) {
3751 return OFPROTO_POSTPONE;
3754 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3755 ? OFPRAW_OFPT10_BARRIER_REPLY
3756 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3757 ofconn_send_reply(ofconn, buf);
3762 ofproto_compose_flow_refresh_update(const struct rule *rule,
3763 enum nx_flow_monitor_flags flags,
3766 struct ofoperation *op = rule->pending;
3767 struct ofputil_flow_update fu;
3770 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3771 /* We'll report the final flow when the operation completes. Reporting
3772 * it now would cause a duplicate report later. */
3776 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3777 ? NXFME_ADDED : NXFME_MODIFIED);
3779 fu.idle_timeout = rule->idle_timeout;
3780 fu.hard_timeout = rule->hard_timeout;
3781 fu.table_id = rule->table_id;
3782 fu.cookie = rule->flow_cookie;
3783 minimatch_expand(&rule->cr.match, &match);
3785 fu.priority = rule->cr.priority;
3786 if (!(flags & NXFMF_ACTIONS)) {
3790 fu.ofpacts = rule->ofpacts;
3791 fu.ofpacts_len = rule->ofpacts_len;
3793 /* An operation is in progress. Use the previous version of the flow's
3794 * actions, so that when the operation commits we report the change. */
3796 case OFOPERATION_ADD:
3797 /* We already verified that there was a victim. */
3798 fu.ofpacts = op->victim->ofpacts;
3799 fu.ofpacts_len = op->victim->ofpacts_len;
3802 case OFOPERATION_MODIFY:
3804 fu.ofpacts = op->ofpacts;
3805 fu.ofpacts_len = op->ofpacts_len;
3807 fu.ofpacts = rule->ofpacts;
3808 fu.ofpacts_len = rule->ofpacts_len;
3812 case OFOPERATION_DELETE:
3813 fu.ofpacts = rule->ofpacts;
3814 fu.ofpacts_len = rule->ofpacts_len;
3822 if (list_is_empty(msgs)) {
3823 ofputil_start_flow_update(msgs);
3825 ofputil_append_flow_update(&fu, msgs);
3829 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3833 LIST_FOR_EACH (rule, ofproto_node, rules) {
3834 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3835 rule->monitor_flags = 0;
3837 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3842 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3843 struct rule *rule, uint64_t seqno,
3846 enum nx_flow_monitor_flags update;
3848 if (ofproto_rule_is_hidden(rule)) {
3853 ? ofoperation_has_out_port(rule->pending, m->out_port)
3854 : ofproto_rule_has_out_port(rule, m->out_port))) {
3859 if (rule->add_seqno > seqno) {
3860 update = NXFMF_ADD | NXFMF_MODIFY;
3861 } else if (rule->modify_seqno > seqno) {
3862 update = NXFMF_MODIFY;
3867 if (!(m->flags & update)) {
3871 update = NXFMF_INITIAL;
3874 if (!rule->monitor_flags) {
3875 list_push_back(rules, &rule->ofproto_node);
3877 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3881 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3885 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3886 const struct ofoperation *op;
3887 const struct oftable *table;
3888 struct cls_rule target;
3890 cls_rule_init_from_minimatch(&target, &m->match, 0);
3891 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3892 struct cls_cursor cursor;
3895 cls_cursor_init(&cursor, &table->cls, &target);
3896 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3897 assert(!rule->pending); /* XXX */
3898 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3902 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3903 struct rule *rule = op->rule;
3905 if (((m->table_id == 0xff
3906 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3907 : m->table_id == rule->table_id))
3908 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3909 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3912 cls_rule_destroy(&target);
3916 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3919 if (m->flags & NXFMF_INITIAL) {
3920 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3925 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3926 uint64_t seqno, struct list *rules)
3928 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3932 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3934 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3935 struct ofmonitor **monitors;
3936 size_t n_monitors, allocated_monitors;
3937 struct list replies;
3944 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3946 n_monitors = allocated_monitors = 0;
3948 struct ofputil_flow_monitor_request request;
3949 struct ofmonitor *m;
3952 retval = ofputil_decode_flow_monitor_request(&request, &b);
3953 if (retval == EOF) {
3955 } else if (retval) {
3960 if (request.table_id != 0xff
3961 && request.table_id >= ofproto->n_tables) {
3962 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3966 error = ofmonitor_create(&request, ofconn, &m);
3971 if (n_monitors >= allocated_monitors) {
3972 monitors = x2nrealloc(monitors, &allocated_monitors,
3975 monitors[n_monitors++] = m;
3979 for (i = 0; i < n_monitors; i++) {
3980 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3983 ofpmp_init(&replies, oh);
3984 ofmonitor_compose_refresh_updates(&rules, &replies);
3985 ofconn_send_replies(ofconn, &replies);
3992 for (i = 0; i < n_monitors; i++) {
3993 ofmonitor_destroy(monitors[i]);
4000 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4002 struct ofmonitor *m;
4005 id = ofputil_decode_flow_monitor_cancel(oh);
4006 m = ofmonitor_lookup(ofconn, id);
4008 return OFPERR_NXBRC_FM_BAD_ID;
4011 ofmonitor_destroy(m);
4016 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4018 const struct ofp_header *oh = msg->data;
4022 error = ofptype_decode(&type, oh);
4028 /* OpenFlow requests. */
4029 case OFPTYPE_ECHO_REQUEST:
4030 return handle_echo_request(ofconn, oh);
4032 case OFPTYPE_FEATURES_REQUEST:
4033 return handle_features_request(ofconn, oh);
4035 case OFPTYPE_GET_CONFIG_REQUEST:
4036 return handle_get_config_request(ofconn, oh);
4038 case OFPTYPE_SET_CONFIG:
4039 return handle_set_config(ofconn, oh);
4041 case OFPTYPE_PACKET_OUT:
4042 return handle_packet_out(ofconn, oh);
4044 case OFPTYPE_PORT_MOD:
4045 return handle_port_mod(ofconn, oh);
4047 case OFPTYPE_FLOW_MOD:
4048 return handle_flow_mod(ofconn, oh);
4050 case OFPTYPE_BARRIER_REQUEST:
4051 return handle_barrier_request(ofconn, oh);
4053 /* OpenFlow replies. */
4054 case OFPTYPE_ECHO_REPLY:
4057 /* Nicira extension requests. */
4058 case OFPTYPE_ROLE_REQUEST:
4059 return handle_role_request(ofconn, oh);
4061 case OFPTYPE_FLOW_MOD_TABLE_ID:
4062 return handle_nxt_flow_mod_table_id(ofconn, oh);
4064 case OFPTYPE_SET_FLOW_FORMAT:
4065 return handle_nxt_set_flow_format(ofconn, oh);
4067 case OFPTYPE_SET_PACKET_IN_FORMAT:
4068 return handle_nxt_set_packet_in_format(ofconn, oh);
4070 case OFPTYPE_SET_CONTROLLER_ID:
4071 return handle_nxt_set_controller_id(ofconn, oh);
4073 case OFPTYPE_FLOW_AGE:
4074 /* Nothing to do. */
4077 case OFPTYPE_FLOW_MONITOR_CANCEL:
4078 return handle_flow_monitor_cancel(ofconn, oh);
4080 case OFPTYPE_SET_ASYNC_CONFIG:
4081 return handle_nxt_set_async_config(ofconn, oh);
4083 /* Statistics requests. */
4084 case OFPTYPE_DESC_STATS_REQUEST:
4085 return handle_desc_stats_request(ofconn, oh);
4087 case OFPTYPE_FLOW_STATS_REQUEST:
4088 return handle_flow_stats_request(ofconn, oh);
4090 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4091 return handle_aggregate_stats_request(ofconn, oh);
4093 case OFPTYPE_TABLE_STATS_REQUEST:
4094 return handle_table_stats_request(ofconn, oh);
4096 case OFPTYPE_PORT_STATS_REQUEST:
4097 return handle_port_stats_request(ofconn, oh);
4099 case OFPTYPE_QUEUE_STATS_REQUEST:
4100 return handle_queue_stats_request(ofconn, oh);
4102 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4103 return handle_port_desc_stats_request(ofconn, oh);
4105 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4106 return handle_flow_monitor_request(ofconn, oh);
4108 /* FIXME: Change the following once they are implemented: */
4109 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4110 case OFPTYPE_GET_ASYNC_REQUEST:
4111 case OFPTYPE_METER_MOD:
4112 case OFPTYPE_GROUP_REQUEST:
4113 case OFPTYPE_GROUP_DESC_REQUEST:
4114 case OFPTYPE_GROUP_FEATURES_REQUEST:
4115 case OFPTYPE_METER_REQUEST:
4116 case OFPTYPE_METER_CONFIG_REQUEST:
4117 case OFPTYPE_METER_FEATURES_REQUEST:
4118 case OFPTYPE_TABLE_FEATURES_REQUEST:
4119 return OFPERR_OFPBRC_BAD_TYPE;
4123 case OFPTYPE_FEATURES_REPLY:
4124 case OFPTYPE_GET_CONFIG_REPLY:
4125 case OFPTYPE_PACKET_IN:
4126 case OFPTYPE_FLOW_REMOVED:
4127 case OFPTYPE_PORT_STATUS:
4128 case OFPTYPE_BARRIER_REPLY:
4129 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4130 case OFPTYPE_DESC_STATS_REPLY:
4131 case OFPTYPE_FLOW_STATS_REPLY:
4132 case OFPTYPE_QUEUE_STATS_REPLY:
4133 case OFPTYPE_PORT_STATS_REPLY:
4134 case OFPTYPE_TABLE_STATS_REPLY:
4135 case OFPTYPE_AGGREGATE_STATS_REPLY:
4136 case OFPTYPE_PORT_DESC_STATS_REPLY:
4137 case OFPTYPE_ROLE_REPLY:
4138 case OFPTYPE_FLOW_MONITOR_PAUSED:
4139 case OFPTYPE_FLOW_MONITOR_RESUMED:
4140 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4141 case OFPTYPE_GET_ASYNC_REPLY:
4142 case OFPTYPE_GROUP_REPLY:
4143 case OFPTYPE_GROUP_DESC_REPLY:
4144 case OFPTYPE_GROUP_FEATURES_REPLY:
4145 case OFPTYPE_METER_REPLY:
4146 case OFPTYPE_METER_CONFIG_REPLY:
4147 case OFPTYPE_METER_FEATURES_REPLY:
4148 case OFPTYPE_TABLE_FEATURES_REPLY:
4150 return OFPERR_OFPBRC_BAD_TYPE;
4155 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
4157 int error = handle_openflow__(ofconn, ofp_msg);
4158 if (error && error != OFPROTO_POSTPONE) {
4159 ofconn_send_error(ofconn, ofp_msg->data, error);
4161 COVERAGE_INC(ofproto_recv_openflow);
4162 return error != OFPROTO_POSTPONE;
4165 /* Asynchronous operations. */
4167 /* Creates and returns a new ofopgroup that is not associated with any
4168 * OpenFlow connection.
4170 * The caller should add operations to the returned group with
4171 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4172 static struct ofopgroup *
4173 ofopgroup_create_unattached(struct ofproto *ofproto)
4175 struct ofopgroup *group = xzalloc(sizeof *group);
4176 group->ofproto = ofproto;
4177 list_init(&group->ofproto_node);
4178 list_init(&group->ops);
4179 list_init(&group->ofconn_node);
4183 /* Creates and returns a new ofopgroup for 'ofproto'.
4185 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4186 * connection. The 'request' and 'buffer_id' arguments are ignored.
4188 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4189 * If the ofopgroup eventually fails, then the error reply will include
4190 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4191 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4193 * The caller should add operations to the returned group with
4194 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4195 static struct ofopgroup *
4196 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4197 const struct ofp_header *request, uint32_t buffer_id)
4199 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4201 size_t request_len = ntohs(request->length);
4203 assert(ofconn_get_ofproto(ofconn) == ofproto);
4205 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4206 group->ofconn = ofconn;
4207 group->request = xmemdup(request, MIN(request_len, 64));
4208 group->buffer_id = buffer_id;
4213 /* Submits 'group' for processing.
4215 * If 'group' contains no operations (e.g. none were ever added, or all of the
4216 * ones that were added completed synchronously), then it is destroyed
4217 * immediately. Otherwise it is added to the ofproto's list of pending
4220 ofopgroup_submit(struct ofopgroup *group)
4222 if (!group->n_running) {
4223 ofopgroup_complete(group);
4225 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4226 group->ofproto->n_pending++;
4231 ofopgroup_complete(struct ofopgroup *group)
4233 struct ofproto *ofproto = group->ofproto;
4235 struct ofconn *abbrev_ofconn;
4236 ovs_be32 abbrev_xid;
4238 struct ofoperation *op, *next_op;
4241 assert(!group->n_running);
4244 LIST_FOR_EACH (op, group_node, &group->ops) {
4251 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4252 LIST_FOR_EACH (op, group_node, &group->ops) {
4253 if (op->type != OFOPERATION_DELETE) {
4254 struct ofpbuf *packet;
4257 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4261 error = rule_execute(op->rule, in_port, packet);
4268 if (!error && !list_is_empty(&group->ofconn_node)) {
4269 abbrev_ofconn = group->ofconn;
4270 abbrev_xid = group->request->xid;
4272 abbrev_ofconn = NULL;
4273 abbrev_xid = htonl(0);
4275 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4276 struct rule *rule = op->rule;
4278 if (!op->error && !ofproto_rule_is_hidden(rule)) {
4279 /* Check that we can just cast from ofoperation_type to
4280 * nx_flow_update_event. */
4281 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4283 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4285 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4288 ofmonitor_report(ofproto->connmgr, rule,
4289 (enum nx_flow_update_event) op->type,
4290 op->reason, abbrev_ofconn, abbrev_xid);
4293 rule->pending = NULL;
4296 case OFOPERATION_ADD:
4300 ofproto_rule_destroy__(op->victim);
4301 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4302 if (vid_mask == VLAN_VID_MASK) {
4303 if (ofproto->vlan_bitmap) {
4304 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4305 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4306 bitmap_set1(ofproto->vlan_bitmap, vid);
4307 ofproto->vlans_changed = true;
4310 ofproto->vlans_changed = true;
4314 oftable_substitute_rule(rule, op->victim);
4315 ofproto_rule_destroy__(rule);
4319 case OFOPERATION_DELETE:
4321 ofproto_rule_destroy__(rule);
4325 case OFOPERATION_MODIFY:
4327 rule->modified = time_msec();
4329 rule->flow_cookie = op->flow_cookie;
4331 free(rule->ofpacts);
4332 rule->ofpacts = op->ofpacts;
4333 rule->ofpacts_len = op->ofpacts_len;
4335 op->ofpacts_len = 0;
4344 ofoperation_destroy(op);
4347 ofmonitor_flush(ofproto->connmgr);
4349 if (!list_is_empty(&group->ofproto_node)) {
4350 assert(ofproto->n_pending > 0);
4351 ofproto->n_pending--;
4352 list_remove(&group->ofproto_node);
4354 if (!list_is_empty(&group->ofconn_node)) {
4355 list_remove(&group->ofconn_node);
4357 ofconn_send_error(group->ofconn, group->request, error);
4359 connmgr_retry(ofproto->connmgr);
4361 free(group->request);
4365 /* Initiates a new operation on 'rule', of the specified 'type', within
4366 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4368 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4369 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4371 * Returns the newly created ofoperation (which is also available as
4372 * rule->pending). */
4373 static struct ofoperation *
4374 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4375 enum ofoperation_type type,
4376 enum ofp_flow_removed_reason reason)
4378 struct ofproto *ofproto = group->ofproto;
4379 struct ofoperation *op;
4381 assert(!rule->pending);
4383 op = rule->pending = xzalloc(sizeof *op);
4385 list_push_back(&group->ops, &op->group_node);
4388 op->reason = reason;
4389 op->flow_cookie = rule->flow_cookie;
4393 if (type == OFOPERATION_DELETE) {
4394 hmap_insert(&ofproto->deletions, &op->hmap_node,
4395 cls_rule_hash(&rule->cr, rule->table_id));
4402 ofoperation_destroy(struct ofoperation *op)
4404 struct ofopgroup *group = op->group;
4407 op->rule->pending = NULL;
4409 if (op->type == OFOPERATION_DELETE) {
4410 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4412 list_remove(&op->group_node);
4417 /* Indicates that 'op' completed with status 'error', which is either 0 to
4418 * indicate success or an OpenFlow error code on failure.
4420 * If 'error' is 0, indicating success, the operation will be committed
4421 * permanently to the flow table. There is one interesting subcase:
4423 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4424 * the flow table (the "victim" rule) by a new one, then the caller must
4425 * have uninitialized any derived state in the victim rule, as in step 5 in
4426 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4427 * performs steps 6 and 7 for the victim rule, most notably by calling its
4428 * ->rule_dealloc() function.
4430 * If 'error' is nonzero, then generally the operation will be rolled back:
4432 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4433 * restores the original rule. The caller must have uninitialized any
4434 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4435 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4436 * and 7 for the new rule, calling its ->rule_dealloc() function.
4438 * - If 'op' is a "modify flow" operation, ofproto restores the original
4441 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4442 * allowed to fail. It must always succeed.
4444 * Please see the large comment in ofproto/ofproto-provider.h titled
4445 * "Asynchronous Operation Support" for more information. */
4447 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4449 struct ofopgroup *group = op->group;
4451 assert(op->rule->pending == op);
4452 assert(group->n_running > 0);
4453 assert(!error || op->type != OFOPERATION_DELETE);
4456 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4457 ofopgroup_complete(group);
4462 ofoperation_get_victim(struct ofoperation *op)
4464 assert(op->type == OFOPERATION_ADD);
4469 pick_datapath_id(const struct ofproto *ofproto)
4471 const struct ofport *port;
4473 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4475 uint8_t ea[ETH_ADDR_LEN];
4478 error = netdev_get_etheraddr(port->netdev, ea);
4480 return eth_addr_to_uint64(ea);
4482 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4483 ofproto->name, netdev_get_name(port->netdev),
4486 return ofproto->fallback_dpid;
4490 pick_fallback_dpid(void)
4492 uint8_t ea[ETH_ADDR_LEN];
4493 eth_addr_nicira_random(ea);
4494 return eth_addr_to_uint64(ea);
4497 /* Table overflow policy. */
4499 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4500 * is not configured to evict rules or if the table contains no evictable
4501 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4503 static struct rule *
4504 choose_rule_to_evict(struct oftable *table)
4506 struct eviction_group *evg;
4508 if (!table->eviction_fields) {
4512 /* In the common case, the outer and inner loops here will each be entered
4515 * - The inner loop normally "return"s in its first iteration. If the
4516 * eviction group has any evictable rules, then it always returns in
4519 * - The outer loop only iterates more than once if the largest eviction
4520 * group has no evictable rules.
4522 * - The outer loop can exit only if table's 'max_flows' is all filled up
4523 * by unevictable rules'. */
4524 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4527 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4528 if (rule->evictable) {
4537 /* Searches 'ofproto' for tables that have more flows than their configured
4538 * maximum and that have flow eviction enabled, and evicts as many flows as
4539 * necessary and currently feasible from them.
4541 * This triggers only when an OpenFlow table has N flows in it and then the
4542 * client configures a maximum number of flows less than N. */
4544 ofproto_evict(struct ofproto *ofproto)
4546 struct ofopgroup *group;
4547 struct oftable *table;
4549 group = ofopgroup_create_unattached(ofproto);
4550 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4551 while (classifier_count(&table->cls) > table->max_flows
4552 && table->eviction_fields) {
4555 rule = choose_rule_to_evict(table);
4556 if (!rule || rule->pending) {
4560 ofoperation_create(group, rule,
4561 OFOPERATION_DELETE, OFPRR_EVICTION);
4562 oftable_remove_rule(rule);
4563 ofproto->ofproto_class->rule_destruct(rule);
4566 ofopgroup_submit(group);
4569 /* Eviction groups. */
4571 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4572 * rules. The priority contains low-order random bits to ensure that eviction
4573 * groups with the same number of rules are prioritized randomly. */
4575 eviction_group_priority(size_t n_rules)
4577 uint16_t size = MIN(UINT16_MAX, n_rules);
4578 return (size << 16) | random_uint16();
4581 /* Updates 'evg', an eviction_group within 'table', following a change that
4582 * adds or removes rules in 'evg'. */
4584 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4586 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4587 eviction_group_priority(heap_count(&evg->rules)));
4590 /* Destroys 'evg', an eviction_group within 'table':
4592 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4593 * rules themselves, just removes them from the eviction group.)
4595 * - Removes 'evg' from 'table'.
4599 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4601 while (!heap_is_empty(&evg->rules)) {
4604 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4605 rule->eviction_group = NULL;
4607 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4608 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4609 heap_destroy(&evg->rules);
4613 /* Removes 'rule' from its eviction group, if any. */
4615 eviction_group_remove_rule(struct rule *rule)
4617 if (rule->eviction_group) {
4618 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4619 struct eviction_group *evg = rule->eviction_group;
4621 rule->eviction_group = NULL;
4622 heap_remove(&evg->rules, &rule->evg_node);
4623 if (heap_is_empty(&evg->rules)) {
4624 eviction_group_destroy(table, evg);
4626 eviction_group_resized(table, evg);
4631 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4632 * returns the hash value. */
4634 eviction_group_hash_rule(struct rule *rule)
4636 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4637 const struct mf_subfield *sf;
4641 hash = table->eviction_group_id_basis;
4642 miniflow_expand(&rule->cr.match.flow, &flow);
4643 for (sf = table->eviction_fields;
4644 sf < &table->eviction_fields[table->n_eviction_fields];
4647 if (mf_are_prereqs_ok(sf->field, &flow)) {
4648 union mf_value value;
4650 mf_get_value(sf->field, &flow, &value);
4652 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4654 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4655 unsigned int start = sf->ofs + sf->n_bits;
4656 bitwise_zero(&value, sf->field->n_bytes, start,
4657 sf->field->n_bytes * 8 - start);
4659 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4661 hash = hash_int(hash, 0);
4668 /* Returns an eviction group within 'table' with the given 'id', creating one
4670 static struct eviction_group *
4671 eviction_group_find(struct oftable *table, uint32_t id)
4673 struct eviction_group *evg;
4675 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4679 evg = xmalloc(sizeof *evg);
4680 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4681 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4682 eviction_group_priority(0));
4683 heap_init(&evg->rules);
4688 /* Returns an eviction priority for 'rule'. The return value should be
4689 * interpreted so that higher priorities make a rule more attractive candidates
4692 rule_eviction_priority(struct rule *rule)
4694 long long int hard_expiration;
4695 long long int idle_expiration;
4696 long long int expiration;
4697 uint32_t expiration_offset;
4699 /* Calculate time of expiration. */
4700 hard_expiration = (rule->hard_timeout
4701 ? rule->modified + rule->hard_timeout * 1000
4703 idle_expiration = (rule->idle_timeout
4704 ? rule->used + rule->idle_timeout * 1000
4706 expiration = MIN(hard_expiration, idle_expiration);
4707 if (expiration == LLONG_MAX) {
4711 /* Calculate the time of expiration as a number of (approximate) seconds
4712 * after program startup.
4714 * This should work OK for program runs that last UINT32_MAX seconds or
4715 * less. Therefore, please restart OVS at least once every 136 years. */
4716 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4718 /* Invert the expiration offset because we're using a max-heap. */
4719 return UINT32_MAX - expiration_offset;
4722 /* Adds 'rule' to an appropriate eviction group for its oftable's
4723 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4724 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4727 * The caller must ensure that 'rule' is not already in an eviction group. */
4729 eviction_group_add_rule(struct rule *rule)
4731 struct ofproto *ofproto = rule->ofproto;
4732 struct oftable *table = &ofproto->tables[rule->table_id];
4734 if (table->eviction_fields
4735 && (rule->hard_timeout || rule->idle_timeout)) {
4736 struct eviction_group *evg;
4738 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4740 rule->eviction_group = evg;
4741 heap_insert(&evg->rules, &rule->evg_node,
4742 rule_eviction_priority(rule));
4743 eviction_group_resized(table, evg);
4749 /* Initializes 'table'. */
4751 oftable_init(struct oftable *table)
4753 memset(table, 0, sizeof *table);
4754 classifier_init(&table->cls);
4755 table->max_flows = UINT_MAX;
4758 /* Destroys 'table', including its classifier and eviction groups.
4760 * The caller is responsible for freeing 'table' itself. */
4762 oftable_destroy(struct oftable *table)
4764 assert(classifier_is_empty(&table->cls));
4765 oftable_disable_eviction(table);
4766 classifier_destroy(&table->cls);
4770 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4771 * string, then 'table' will use its default name.
4773 * This only affects the name exposed for a table exposed through the OpenFlow
4774 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4776 oftable_set_name(struct oftable *table, const char *name)
4778 if (name && name[0]) {
4779 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4780 if (!table->name || strncmp(name, table->name, len)) {
4782 table->name = xmemdup0(name, len);
4790 /* oftables support a choice of two policies when adding a rule would cause the
4791 * number of flows in the table to exceed the configured maximum number: either
4792 * they can refuse to add the new flow or they can evict some existing flow.
4793 * This function configures the former policy on 'table'. */
4795 oftable_disable_eviction(struct oftable *table)
4797 if (table->eviction_fields) {
4798 struct eviction_group *evg, *next;
4800 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4801 &table->eviction_groups_by_id) {
4802 eviction_group_destroy(table, evg);
4804 hmap_destroy(&table->eviction_groups_by_id);
4805 heap_destroy(&table->eviction_groups_by_size);
4807 free(table->eviction_fields);
4808 table->eviction_fields = NULL;
4809 table->n_eviction_fields = 0;
4813 /* oftables support a choice of two policies when adding a rule would cause the
4814 * number of flows in the table to exceed the configured maximum number: either
4815 * they can refuse to add the new flow or they can evict some existing flow.
4816 * This function configures the latter policy on 'table', with fairness based
4817 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4818 * 'n_fields' as 0 disables fairness.) */
4820 oftable_enable_eviction(struct oftable *table,
4821 const struct mf_subfield *fields, size_t n_fields)
4823 struct cls_cursor cursor;
4826 if (table->eviction_fields
4827 && n_fields == table->n_eviction_fields
4829 || !memcmp(fields, table->eviction_fields,
4830 n_fields * sizeof *fields))) {
4835 oftable_disable_eviction(table);
4837 table->n_eviction_fields = n_fields;
4838 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4840 table->eviction_group_id_basis = random_uint32();
4841 hmap_init(&table->eviction_groups_by_id);
4842 heap_init(&table->eviction_groups_by_size);
4844 cls_cursor_init(&cursor, &table->cls, NULL);
4845 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4846 eviction_group_add_rule(rule);
4850 /* Removes 'rule' from the oftable that contains it. */
4852 oftable_remove_rule(struct rule *rule)
4854 struct ofproto *ofproto = rule->ofproto;
4855 struct oftable *table = &ofproto->tables[rule->table_id];
4857 classifier_remove(&table->cls, &rule->cr);
4858 eviction_group_remove_rule(rule);
4861 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4862 * oftable that has an identical cls_rule. Returns the rule that was removed,
4863 * if any, and otherwise NULL. */
4864 static struct rule *
4865 oftable_replace_rule(struct rule *rule)
4867 struct ofproto *ofproto = rule->ofproto;
4868 struct oftable *table = &ofproto->tables[rule->table_id];
4869 struct rule *victim;
4871 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4873 eviction_group_remove_rule(victim);
4875 eviction_group_add_rule(rule);
4879 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4881 oftable_substitute_rule(struct rule *old, struct rule *new)
4884 oftable_replace_rule(new);
4886 oftable_remove_rule(old);
4890 /* unixctl commands. */
4893 ofproto_lookup(const char *name)
4895 struct ofproto *ofproto;
4897 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4899 if (!strcmp(ofproto->name, name)) {
4907 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4908 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4910 struct ofproto *ofproto;
4914 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4915 ds_put_format(&results, "%s\n", ofproto->name);
4917 unixctl_command_reply(conn, ds_cstr(&results));
4918 ds_destroy(&results);
4922 ofproto_unixctl_init(void)
4924 static bool registered;
4930 unixctl_command_register("ofproto/list", "", 0, 0,
4931 ofproto_unixctl_list, NULL);
4934 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4936 * This is deprecated. It is only for compatibility with broken device drivers
4937 * in old versions of Linux that do not properly support VLANs when VLAN
4938 * devices are not used. When broken device drivers are no longer in
4939 * widespread use, we will delete these interfaces. */
4941 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4942 * (exactly) by an OpenFlow rule in 'ofproto'. */
4944 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4946 const struct oftable *oftable;
4948 free(ofproto->vlan_bitmap);
4949 ofproto->vlan_bitmap = bitmap_allocate(4096);
4950 ofproto->vlans_changed = false;
4952 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4953 const struct cls_table *table;
4955 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4956 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4957 const struct cls_rule *rule;
4959 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4960 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4961 bitmap_set1(vlan_bitmap, vid);
4962 bitmap_set1(ofproto->vlan_bitmap, vid);
4969 /* Returns true if new VLANs have come into use by the flow table since the
4970 * last call to ofproto_get_vlan_usage().
4972 * We don't track when old VLANs stop being used. */
4974 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4976 return ofproto->vlans_changed;
4979 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4980 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4981 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4982 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4983 * then the VLAN device is un-enslaved. */
4985 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4986 uint16_t realdev_ofp_port, int vid)
4988 struct ofport *ofport;
4991 assert(vlandev_ofp_port != realdev_ofp_port);
4993 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4995 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4996 ofproto->name, vlandev_ofp_port);
5000 if (!ofproto->ofproto_class->set_realdev) {
5001 if (!vlandev_ofp_port) {
5004 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5008 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5010 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5011 ofproto->name, vlandev_ofp_port,
5012 netdev_get_name(ofport->netdev), strerror(error));