2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
206 static void calc_duration(long long int start, long long int now,
207 uint32_t *sec, uint32_t *nsec);
210 static uint64_t pick_datapath_id(const struct ofproto *);
211 static uint64_t pick_fallback_dpid(void);
212 static void ofproto_destroy__(struct ofproto *);
213 static void update_mtu(struct ofproto *, struct ofport *);
216 static void ofproto_unixctl_init(void);
218 /* All registered ofproto classes, in probe order. */
219 static const struct ofproto_class **ofproto_classes;
220 static size_t n_ofproto_classes;
221 static size_t allocated_ofproto_classes;
223 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
225 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
226 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
228 /* Initial mappings of port to OpenFlow number mappings. */
229 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
231 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
233 /* The default value of true waits for flow restore. */
234 static bool flow_restore_wait = true;
236 /* Must be called to initialize the ofproto library.
238 * The caller may pass in 'iface_hints', which contains an shash of
239 * "iface_hint" elements indexed by the interface's name. The provider
240 * may use these hints to describe the startup configuration in order to
241 * reinitialize its state. The caller owns the provided data, so a
242 * provider will make copies of anything required. An ofproto provider
243 * will remove any existing state that is not described by the hint, and
244 * may choose to remove it all. */
246 ofproto_init(const struct shash *iface_hints)
248 struct shash_node *node;
251 ofproto_class_register(&ofproto_dpif_class);
253 /* Make a local copy, since we don't own 'iface_hints' elements. */
254 SHASH_FOR_EACH(node, iface_hints) {
255 const struct iface_hint *orig_hint = node->data;
256 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
257 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
259 new_hint->br_name = xstrdup(orig_hint->br_name);
260 new_hint->br_type = xstrdup(br_type);
261 new_hint->ofp_port = orig_hint->ofp_port;
263 shash_add(&init_ofp_ports, node->name, new_hint);
266 for (i = 0; i < n_ofproto_classes; i++) {
267 ofproto_classes[i]->init(&init_ofp_ports);
271 /* 'type' should be a normalized datapath type, as returned by
272 * ofproto_normalize_type(). Returns the corresponding ofproto_class
273 * structure, or a null pointer if there is none registered for 'type'. */
274 static const struct ofproto_class *
275 ofproto_class_find__(const char *type)
279 for (i = 0; i < n_ofproto_classes; i++) {
280 const struct ofproto_class *class = ofproto_classes[i];
285 class->enumerate_types(&types);
286 found = sset_contains(&types, type);
287 sset_destroy(&types);
293 VLOG_WARN("unknown datapath type %s", type);
297 /* Registers a new ofproto class. After successful registration, new ofprotos
298 * of that type can be created using ofproto_create(). */
300 ofproto_class_register(const struct ofproto_class *new_class)
304 for (i = 0; i < n_ofproto_classes; i++) {
305 if (ofproto_classes[i] == new_class) {
310 if (n_ofproto_classes >= allocated_ofproto_classes) {
311 ofproto_classes = x2nrealloc(ofproto_classes,
312 &allocated_ofproto_classes,
313 sizeof *ofproto_classes);
315 ofproto_classes[n_ofproto_classes++] = new_class;
319 /* Unregisters a datapath provider. 'type' must have been previously
320 * registered and not currently be in use by any ofprotos. After
321 * unregistration new datapaths of that type cannot be opened using
322 * ofproto_create(). */
324 ofproto_class_unregister(const struct ofproto_class *class)
328 for (i = 0; i < n_ofproto_classes; i++) {
329 if (ofproto_classes[i] == class) {
330 for (i++; i < n_ofproto_classes; i++) {
331 ofproto_classes[i - 1] = ofproto_classes[i];
337 VLOG_WARN("attempted to unregister an ofproto class that is not "
342 /* Clears 'types' and enumerates all registered ofproto types into it. The
343 * caller must first initialize the sset. */
345 ofproto_enumerate_types(struct sset *types)
349 for (i = 0; i < n_ofproto_classes; i++) {
350 ofproto_classes[i]->enumerate_types(types);
354 /* Returns the fully spelled out name for the given ofproto 'type'.
356 * Normalized type string can be compared with strcmp(). Unnormalized type
357 * string might be the same even if they have different spellings. */
359 ofproto_normalize_type(const char *type)
361 return type && type[0] ? type : "system";
364 /* Clears 'names' and enumerates the names of all known created ofprotos with
365 * the given 'type'. The caller must first initialize the sset. Returns 0 if
366 * successful, otherwise a positive errno value.
368 * Some kinds of datapaths might not be practically enumerable. This is not
369 * considered an error. */
371 ofproto_enumerate_names(const char *type, struct sset *names)
373 const struct ofproto_class *class = ofproto_class_find__(type);
374 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
378 ofproto_create(const char *datapath_name, const char *datapath_type,
379 struct ofproto **ofprotop)
381 const struct ofproto_class *class;
382 struct ofproto *ofproto;
388 ofproto_unixctl_init();
390 datapath_type = ofproto_normalize_type(datapath_type);
391 class = ofproto_class_find__(datapath_type);
393 VLOG_WARN("could not create datapath %s of unknown type %s",
394 datapath_name, datapath_type);
398 ofproto = class->alloc();
400 VLOG_ERR("failed to allocate datapath %s of type %s",
401 datapath_name, datapath_type);
406 memset(ofproto, 0, sizeof *ofproto);
407 ofproto->ofproto_class = class;
408 ofproto->name = xstrdup(datapath_name);
409 ofproto->type = xstrdup(datapath_type);
410 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
411 hash_string(ofproto->name, 0));
412 ofproto->datapath_id = 0;
413 ofproto->forward_bpdu = false;
414 ofproto->fallback_dpid = pick_fallback_dpid();
415 ofproto->mfr_desc = NULL;
416 ofproto->hw_desc = NULL;
417 ofproto->sw_desc = NULL;
418 ofproto->serial_desc = NULL;
419 ofproto->dp_desc = NULL;
420 ofproto->frag_handling = OFPC_FRAG_NORMAL;
421 hmap_init(&ofproto->ports);
422 shash_init(&ofproto->port_by_name);
423 simap_init(&ofproto->ofp_requests);
424 ofproto->max_ports = OFPP_MAX;
425 ofproto->tables = NULL;
426 ofproto->n_tables = 0;
427 list_init(&ofproto->expirable);
428 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
429 ofproto->state = S_OPENFLOW;
430 list_init(&ofproto->pending);
431 ofproto->n_pending = 0;
432 hmap_init(&ofproto->deletions);
433 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
434 ofproto->first_op = ofproto->last_op = LLONG_MIN;
435 ofproto->next_op_report = LLONG_MAX;
436 ofproto->op_backoff = LLONG_MIN;
437 ofproto->vlan_bitmap = NULL;
438 ofproto->vlans_changed = false;
439 ofproto->min_mtu = INT_MAX;
441 error = ofproto->ofproto_class->construct(ofproto);
443 VLOG_ERR("failed to open datapath %s: %s",
444 datapath_name, strerror(error));
445 ofproto_destroy__(ofproto);
449 /* The "max_ports" member should have been set by ->construct(ofproto).
450 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
451 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
452 bitmap_set1(ofproto->ofp_port_ids, 0);
454 /* Check that hidden tables, if any, are at the end. */
455 ovs_assert(ofproto->n_tables);
456 for (i = 0; i + 1 < ofproto->n_tables; i++) {
457 enum oftable_flags flags = ofproto->tables[i].flags;
458 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
460 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
463 ofproto->datapath_id = pick_datapath_id(ofproto);
470 /* Must be called (only) by an ofproto implementation in its constructor
471 * function. See the large comment on 'construct' in struct ofproto_class for
474 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
476 struct oftable *table;
478 ovs_assert(!ofproto->n_tables);
479 ovs_assert(n_tables >= 1 && n_tables <= 255);
481 ofproto->n_tables = n_tables;
482 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
483 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
488 /* To be optionally called (only) by an ofproto implementation in its
489 * constructor function. See the large comment on 'construct' in struct
490 * ofproto_class for details.
492 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
493 * will then ensure that actions passed into the ofproto implementation will
494 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
495 * function is not called, there will be no such restriction.
497 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
498 * the 'max_ports' restriction. */
500 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
502 ovs_assert(max_ports <= OFPP_MAX);
503 ofproto->max_ports = max_ports;
507 ofproto_get_datapath_id(const struct ofproto *ofproto)
509 return ofproto->datapath_id;
513 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
515 uint64_t old_dpid = p->datapath_id;
516 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
517 if (p->datapath_id != old_dpid) {
518 /* Force all active connections to reconnect, since there is no way to
519 * notify a controller that the datapath ID has changed. */
520 ofproto_reconnect_controllers(p);
525 ofproto_set_controllers(struct ofproto *p,
526 const struct ofproto_controller *controllers,
527 size_t n_controllers, uint32_t allowed_versions)
529 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
534 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
536 connmgr_set_fail_mode(p->connmgr, fail_mode);
539 /* Drops the connections between 'ofproto' and all of its controllers, forcing
540 * them to reconnect. */
542 ofproto_reconnect_controllers(struct ofproto *ofproto)
544 connmgr_reconnect(ofproto->connmgr);
547 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
548 * in-band control should guarantee access, in the same way that in-band
549 * control guarantees access to OpenFlow controllers. */
551 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
552 const struct sockaddr_in *extras, size_t n)
554 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
557 /* Sets the OpenFlow queue used by flows set up by in-band control on
558 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
559 * flows will use the default queue. */
561 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
563 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
566 /* Sets the number of flows at which eviction from the kernel flow table
569 ofproto_set_flow_eviction_threshold(unsigned threshold)
571 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
575 /* If forward_bpdu is true, the NORMAL action will forward frames with
576 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
577 * the NORMAL action will drop these frames. */
579 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
581 bool old_val = ofproto->forward_bpdu;
582 ofproto->forward_bpdu = forward_bpdu;
583 if (old_val != ofproto->forward_bpdu) {
584 if (ofproto->ofproto_class->forward_bpdu_changed) {
585 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
590 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
591 * 'idle_time', in seconds, and the maximum number of MAC table entries to
594 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
597 if (ofproto->ofproto_class->set_mac_table_config) {
598 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
604 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
607 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
611 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
613 return connmgr_set_snoops(ofproto->connmgr, snoops);
617 ofproto_set_netflow(struct ofproto *ofproto,
618 const struct netflow_options *nf_options)
620 if (nf_options && sset_is_empty(&nf_options->collectors)) {
624 if (ofproto->ofproto_class->set_netflow) {
625 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
627 return nf_options ? EOPNOTSUPP : 0;
632 ofproto_set_sflow(struct ofproto *ofproto,
633 const struct ofproto_sflow_options *oso)
635 if (oso && sset_is_empty(&oso->targets)) {
639 if (ofproto->ofproto_class->set_sflow) {
640 return ofproto->ofproto_class->set_sflow(ofproto, oso);
642 return oso ? EOPNOTSUPP : 0;
647 ofproto_set_ipfix(struct ofproto *ofproto,
648 const struct ofproto_ipfix_bridge_exporter_options *bo,
649 const struct ofproto_ipfix_flow_exporter_options *fo,
652 if (ofproto->ofproto_class->set_ipfix) {
653 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
655 return (bo || fo) ? EOPNOTSUPP : 0;
660 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
662 flow_restore_wait = flow_restore_wait_db;
666 ofproto_get_flow_restore_wait(void)
668 return flow_restore_wait;
672 /* Spanning Tree Protocol (STP) configuration. */
674 /* Configures STP on 'ofproto' using the settings defined in 's'. If
675 * 's' is NULL, disables STP.
677 * Returns 0 if successful, otherwise a positive errno value. */
679 ofproto_set_stp(struct ofproto *ofproto,
680 const struct ofproto_stp_settings *s)
682 return (ofproto->ofproto_class->set_stp
683 ? ofproto->ofproto_class->set_stp(ofproto, s)
687 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
688 * 'enabled' member of 's' is false, then the other members are not
691 * Returns 0 if successful, otherwise a positive errno value. */
693 ofproto_get_stp_status(struct ofproto *ofproto,
694 struct ofproto_stp_status *s)
696 return (ofproto->ofproto_class->get_stp_status
697 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
701 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
702 * in 's'. The caller is responsible for assigning STP port numbers
703 * (using the 'port_num' member in the range of 1 through 255, inclusive)
704 * and ensuring there are no duplicates. If the 's' is NULL, then STP
705 * is disabled on the port.
707 * Returns 0 if successful, otherwise a positive errno value.*/
709 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
710 const struct ofproto_port_stp_settings *s)
712 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
714 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
715 ofproto->name, ofp_port);
719 return (ofproto->ofproto_class->set_stp_port
720 ? ofproto->ofproto_class->set_stp_port(ofport, s)
724 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
725 * 's'. If the 'enabled' member in 's' is false, then the other members
726 * are not meaningful.
728 * Returns 0 if successful, otherwise a positive errno value.*/
730 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
731 struct ofproto_port_stp_status *s)
733 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
735 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
736 "port %"PRIu16, ofproto->name, ofp_port);
740 return (ofproto->ofproto_class->get_stp_port_status
741 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
745 /* Queue DSCP configuration. */
747 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
748 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
749 * to implement QoS. Instead, it is used to implement features which require
750 * knowledge of what queues exist on a port, and some basic information about
753 * Returns 0 if successful, otherwise a positive errno value. */
755 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
756 const struct ofproto_port_queue *queues,
759 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
762 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
763 ofproto->name, ofp_port);
767 return (ofproto->ofproto_class->set_queues
768 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
772 /* Connectivity Fault Management configuration. */
774 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
776 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
778 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
779 if (ofport && ofproto->ofproto_class->set_cfm) {
780 ofproto->ofproto_class->set_cfm(ofport, NULL);
784 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
785 * basic configuration from the configuration members in 'cfm', and the remote
786 * maintenance point ID from remote_mpid. Ignores the statistics members of
789 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
791 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
792 const struct cfm_settings *s)
794 struct ofport *ofport;
797 ofport = ofproto_get_port(ofproto, ofp_port);
799 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
800 ofproto->name, ofp_port);
804 /* XXX: For configuration simplicity, we only support one remote_mpid
805 * outside of the CFM module. It's not clear if this is the correct long
806 * term solution or not. */
807 error = (ofproto->ofproto_class->set_cfm
808 ? ofproto->ofproto_class->set_cfm(ofport, s)
811 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
812 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
817 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
818 * 'ofproto' does not have a port 'ofp_port'. */
820 ofproto_port_set_bfd(struct ofproto *ofproto, uint16_t ofp_port,
821 const struct smap *cfg)
823 struct ofport *ofport;
826 ofport = ofproto_get_port(ofproto, ofp_port);
828 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
829 ofproto->name, ofp_port);
833 error = (ofproto->ofproto_class->set_bfd
834 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
837 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
838 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
843 /* Populates 'status' with key value pairs indicating the status of the BFD
844 * session on 'ofp_port'. This information is intended to be populated in the
845 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
848 ofproto_port_get_bfd_status(struct ofproto *ofproto, uint16_t ofp_port,
851 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
852 return (ofport && ofproto->ofproto_class->get_bfd_status
853 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
857 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
858 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
859 * 0 if LACP partner information is not current (generally indicating a
860 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
862 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
864 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
865 return (ofport && ofproto->ofproto_class->port_is_lacp_current
866 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
872 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
873 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
874 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
875 * configuration plus, if there is more than one slave, a bonding
878 * If 'aux' is already registered then this function updates its configuration
879 * to 's'. Otherwise, this function registers a new bundle.
881 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
884 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
885 const struct ofproto_bundle_settings *s)
887 return (ofproto->ofproto_class->bundle_set
888 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
892 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
893 * If no such bundle has been registered, this has no effect. */
895 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
897 return ofproto_bundle_register(ofproto, aux, NULL);
901 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
902 * If 'aux' is already registered then this function updates its configuration
903 * to 's'. Otherwise, this function registers a new mirror. */
905 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
906 const struct ofproto_mirror_settings *s)
908 return (ofproto->ofproto_class->mirror_set
909 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
913 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
914 * If no mirror has been registered, this has no effect. */
916 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
918 return ofproto_mirror_register(ofproto, aux, NULL);
921 /* Retrieves statistics from mirror associated with client data pointer
922 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
923 * 'bytes', respectively. If a particular counters is not supported,
924 * the appropriate argument is set to UINT64_MAX. */
926 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
927 uint64_t *packets, uint64_t *bytes)
929 if (!ofproto->ofproto_class->mirror_get_stats) {
930 *packets = *bytes = UINT64_MAX;
934 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
938 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
939 * which all packets are flooded, instead of using MAC learning. If
940 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
942 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
945 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
947 return (ofproto->ofproto_class->set_flood_vlans
948 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
952 /* Returns true if 'aux' is a registered bundle that is currently in use as the
953 * output for a mirror. */
955 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
957 return (ofproto->ofproto_class->is_mirror_output_bundle
958 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
962 /* Configuration of OpenFlow tables. */
964 /* Returns the number of OpenFlow tables in 'ofproto'. */
966 ofproto_get_n_tables(const struct ofproto *ofproto)
968 return ofproto->n_tables;
971 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
972 * settings from 's'. 'table_id' must be in the range 0 through the number of
973 * OpenFlow tables in 'ofproto' minus 1, inclusive.
975 * For read-only tables, only the name may be configured. */
977 ofproto_configure_table(struct ofproto *ofproto, int table_id,
978 const struct ofproto_table_settings *s)
980 struct oftable *table;
982 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
983 table = &ofproto->tables[table_id];
985 oftable_set_name(table, s->name);
987 if (table->flags & OFTABLE_READONLY) {
992 oftable_enable_eviction(table, s->groups, s->n_groups);
994 oftable_disable_eviction(table);
997 table->max_flows = s->max_flows;
998 if (classifier_count(&table->cls) > table->max_flows
999 && table->eviction_fields) {
1000 /* 'table' contains more flows than allowed. We might not be able to
1001 * evict them right away because of the asynchronous nature of flow
1002 * table changes. Schedule eviction for later. */
1003 switch (ofproto->state) {
1005 ofproto->state = S_EVICT;
1009 /* We're already deleting flows, nothing more to do. */
1016 ofproto_has_snoops(const struct ofproto *ofproto)
1018 return connmgr_has_snoops(ofproto->connmgr);
1022 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1024 connmgr_get_snoops(ofproto->connmgr, snoops);
1028 ofproto_flush__(struct ofproto *ofproto)
1030 struct ofopgroup *group;
1031 struct oftable *table;
1033 if (ofproto->ofproto_class->flush) {
1034 ofproto->ofproto_class->flush(ofproto);
1037 group = ofopgroup_create_unattached(ofproto);
1038 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1039 struct rule *rule, *next_rule;
1040 struct cls_cursor cursor;
1042 if (table->flags & OFTABLE_HIDDEN) {
1046 cls_cursor_init(&cursor, &table->cls, NULL);
1047 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1048 if (!rule->pending) {
1049 ofoperation_create(group, rule, OFOPERATION_DELETE,
1051 oftable_remove_rule(rule);
1052 ofproto->ofproto_class->rule_destruct(rule);
1056 ofopgroup_submit(group);
1060 ofproto_destroy__(struct ofproto *ofproto)
1062 struct oftable *table;
1064 ovs_assert(list_is_empty(&ofproto->pending));
1065 ovs_assert(!ofproto->n_pending);
1067 connmgr_destroy(ofproto->connmgr);
1069 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1070 free(ofproto->name);
1071 free(ofproto->type);
1072 free(ofproto->mfr_desc);
1073 free(ofproto->hw_desc);
1074 free(ofproto->sw_desc);
1075 free(ofproto->serial_desc);
1076 free(ofproto->dp_desc);
1077 hmap_destroy(&ofproto->ports);
1078 shash_destroy(&ofproto->port_by_name);
1079 bitmap_free(ofproto->ofp_port_ids);
1080 simap_destroy(&ofproto->ofp_requests);
1082 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1083 oftable_destroy(table);
1085 free(ofproto->tables);
1087 hmap_destroy(&ofproto->deletions);
1089 free(ofproto->vlan_bitmap);
1091 ofproto->ofproto_class->dealloc(ofproto);
1095 ofproto_destroy(struct ofproto *p)
1097 struct ofport *ofport, *next_ofport;
1104 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1105 ofport_destroy(ofport);
1108 p->ofproto_class->destruct(p);
1109 ofproto_destroy__(p);
1112 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1113 * kernel datapath, for example, this destroys the datapath in the kernel, and
1114 * with the netdev-based datapath, it tears down the data structures that
1115 * represent the datapath.
1117 * The datapath should not be currently open as an ofproto. */
1119 ofproto_delete(const char *name, const char *type)
1121 const struct ofproto_class *class = ofproto_class_find__(type);
1122 return (!class ? EAFNOSUPPORT
1123 : !class->del ? EACCES
1124 : class->del(type, name));
1128 process_port_change(struct ofproto *ofproto, int error, char *devname)
1130 if (error == ENOBUFS) {
1131 reinit_ports(ofproto);
1132 } else if (!error) {
1133 update_port(ofproto, devname);
1139 ofproto_type_run(const char *datapath_type)
1141 const struct ofproto_class *class;
1144 datapath_type = ofproto_normalize_type(datapath_type);
1145 class = ofproto_class_find__(datapath_type);
1147 error = class->type_run ? class->type_run(datapath_type) : 0;
1148 if (error && error != EAGAIN) {
1149 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1150 datapath_type, strerror(error));
1156 ofproto_type_run_fast(const char *datapath_type)
1158 const struct ofproto_class *class;
1161 datapath_type = ofproto_normalize_type(datapath_type);
1162 class = ofproto_class_find__(datapath_type);
1164 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1165 if (error && error != EAGAIN) {
1166 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1167 datapath_type, strerror(error));
1173 ofproto_type_wait(const char *datapath_type)
1175 const struct ofproto_class *class;
1177 datapath_type = ofproto_normalize_type(datapath_type);
1178 class = ofproto_class_find__(datapath_type);
1180 if (class->type_wait) {
1181 class->type_wait(datapath_type);
1186 ofproto_run(struct ofproto *p)
1188 struct sset changed_netdevs;
1189 const char *changed_netdev;
1190 struct ofport *ofport;
1193 error = p->ofproto_class->run(p);
1194 if (error && error != EAGAIN) {
1195 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1198 if (p->ofproto_class->port_poll) {
1201 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1202 process_port_change(p, error, devname);
1206 /* Update OpenFlow port status for any port whose netdev has changed.
1208 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1209 * destroyed, so it's not safe to update ports directly from the
1210 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1211 * need this two-phase approach. */
1212 sset_init(&changed_netdevs);
1213 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1214 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1215 if (ofport->change_seq != change_seq) {
1216 ofport->change_seq = change_seq;
1217 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1220 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1221 update_port(p, changed_netdev);
1223 sset_destroy(&changed_netdevs);
1227 connmgr_run(p->connmgr, handle_openflow);
1231 connmgr_run(p->connmgr, NULL);
1233 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1234 p->state = S_OPENFLOW;
1239 connmgr_run(p->connmgr, NULL);
1241 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1242 connmgr_flushed(p->connmgr);
1243 p->state = S_OPENFLOW;
1251 if (time_msec() >= p->next_op_report) {
1252 long long int ago = (time_msec() - p->first_op) / 1000;
1253 long long int interval = (p->last_op - p->first_op) / 1000;
1257 ds_put_format(&s, "%d flow_mods ",
1258 p->n_add + p->n_delete + p->n_modify);
1259 if (interval == ago) {
1260 ds_put_format(&s, "in the last %lld s", ago);
1261 } else if (interval) {
1262 ds_put_format(&s, "in the %lld s starting %lld s ago",
1265 ds_put_format(&s, "%lld s ago", ago);
1268 ds_put_cstr(&s, " (");
1270 ds_put_format(&s, "%d adds, ", p->n_add);
1273 ds_put_format(&s, "%d deletes, ", p->n_delete);
1276 ds_put_format(&s, "%d modifications, ", p->n_modify);
1279 ds_put_char(&s, ')');
1281 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1284 p->n_add = p->n_delete = p->n_modify = 0;
1285 p->next_op_report = LLONG_MAX;
1291 /* Performs periodic activity required by 'ofproto' that needs to be done
1292 * with the least possible latency.
1294 * It makes sense to call this function a couple of times per poll loop, to
1295 * provide a significant performance boost on some benchmarks with the
1296 * ofproto-dpif implementation. */
1298 ofproto_run_fast(struct ofproto *p)
1302 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1303 if (error && error != EAGAIN) {
1304 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1305 p->name, strerror(error));
1311 ofproto_wait(struct ofproto *p)
1313 struct ofport *ofport;
1315 p->ofproto_class->wait(p);
1316 if (p->ofproto_class->port_poll_wait) {
1317 p->ofproto_class->port_poll_wait(p);
1320 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1321 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1322 poll_immediate_wake();
1328 connmgr_wait(p->connmgr, true);
1333 connmgr_wait(p->connmgr, false);
1334 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1335 poll_immediate_wake();
1342 ofproto_is_alive(const struct ofproto *p)
1344 return connmgr_has_controllers(p->connmgr);
1347 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1348 * memory_report(). */
1350 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1352 const struct oftable *table;
1353 unsigned int n_rules;
1355 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1356 simap_increase(usage, "ops",
1357 ofproto->n_pending + hmap_count(&ofproto->deletions));
1360 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1361 n_rules += classifier_count(&table->cls);
1363 simap_increase(usage, "rules", n_rules);
1365 if (ofproto->ofproto_class->get_memory_usage) {
1366 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1369 connmgr_get_memory_usage(ofproto->connmgr, usage);
1373 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1376 connmgr_get_controller_info(ofproto->connmgr, info);
1380 ofproto_free_ofproto_controller_info(struct shash *info)
1382 connmgr_free_controller_info(info);
1385 /* Makes a deep copy of 'old' into 'port'. */
1387 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1389 port->name = xstrdup(old->name);
1390 port->type = xstrdup(old->type);
1391 port->ofp_port = old->ofp_port;
1394 /* Frees memory allocated to members of 'ofproto_port'.
1396 * Do not call this function on an ofproto_port obtained from
1397 * ofproto_port_dump_next(): that function retains ownership of the data in the
1400 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1402 free(ofproto_port->name);
1403 free(ofproto_port->type);
1406 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1408 * This function provides no status indication. An error status for the entire
1409 * dump operation is provided when it is completed by calling
1410 * ofproto_port_dump_done().
1413 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1414 const struct ofproto *ofproto)
1416 dump->ofproto = ofproto;
1417 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1421 /* Attempts to retrieve another port from 'dump', which must have been created
1422 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1423 * 'port' and returns true. On failure, returns false.
1425 * Failure might indicate an actual error or merely that the last port has been
1426 * dumped. An error status for the entire dump operation is provided when it
1427 * is completed by calling ofproto_port_dump_done().
1429 * The ofproto owns the data stored in 'port'. It will remain valid until at
1430 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1431 * ofproto_port_dump_done(). */
1433 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1434 struct ofproto_port *port)
1436 const struct ofproto *ofproto = dump->ofproto;
1442 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1445 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1451 /* Completes port table dump operation 'dump', which must have been created
1452 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1453 * error-free, otherwise a positive errno value describing the problem. */
1455 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1457 const struct ofproto *ofproto = dump->ofproto;
1459 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1462 return dump->error == EOF ? 0 : dump->error;
1465 /* Returns the type to pass to netdev_open() when a datapath of type
1466 * 'datapath_type' has a port of type 'port_type', for a few special
1467 * cases when a netdev type differs from a port type. For example, when
1468 * using the userspace datapath, a port of type "internal" needs to be
1471 * Returns either 'type' itself or a string literal, which must not be
1474 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1476 const struct ofproto_class *class;
1478 datapath_type = ofproto_normalize_type(datapath_type);
1479 class = ofproto_class_find__(datapath_type);
1484 return (class->port_open_type
1485 ? class->port_open_type(datapath_type, port_type)
1489 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1490 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1491 * the port's OpenFlow port number.
1493 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1494 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1495 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1496 * 'ofp_portp' is non-null). */
1498 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1499 uint16_t *ofp_portp)
1501 uint16_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1504 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1506 const char *netdev_name = netdev_get_name(netdev);
1508 simap_put(&ofproto->ofp_requests, netdev_name, ofp_port);
1509 update_port(ofproto, netdev_name);
1512 struct ofproto_port ofproto_port;
1514 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1516 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1517 ofproto_port_destroy(&ofproto_port);
1522 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1523 * initializes '*port' appropriately; on failure, returns a positive errno
1526 * The caller owns the data in 'ofproto_port' and must free it with
1527 * ofproto_port_destroy() when it is no longer needed. */
1529 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1530 struct ofproto_port *port)
1534 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1536 memset(port, 0, sizeof *port);
1541 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1542 * Returns 0 if successful, otherwise a positive errno. */
1544 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1546 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1547 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1548 struct simap_node *ofp_request_node;
1551 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1552 if (ofp_request_node) {
1553 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1556 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1557 if (!error && ofport) {
1558 /* 'name' is the netdev's name and update_port() is going to close the
1559 * netdev. Just in case update_port() refers to 'name' after it
1560 * destroys 'ofport', make a copy of it around the update_port()
1562 char *devname = xstrdup(name);
1563 update_port(ofproto, devname);
1569 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1570 * performs the 'n_actions' actions in 'actions'. The new flow will not
1573 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1574 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1575 * controllers; otherwise, it will be hidden.
1577 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1579 * This is a helper function for in-band control and fail-open. */
1581 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1582 unsigned int priority,
1583 const struct ofpact *ofpacts, size_t ofpacts_len)
1585 const struct rule *rule;
1587 rule = rule_from_cls_rule(classifier_find_match_exactly(
1588 &ofproto->tables[0].cls, match, priority));
1589 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1590 ofpacts, ofpacts_len)) {
1591 struct ofputil_flow_mod fm;
1593 memset(&fm, 0, sizeof fm);
1595 fm.priority = priority;
1596 fm.buffer_id = UINT32_MAX;
1597 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1598 fm.ofpacts_len = ofpacts_len;
1599 add_flow(ofproto, NULL, &fm, NULL);
1604 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1605 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1606 * operation cannot be initiated now but may be retried later.
1608 * This is a helper function for in-band control and fail-open. */
1610 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1612 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1615 /* Searches for a rule with matching criteria exactly equal to 'target' in
1616 * ofproto's table 0 and, if it finds one, deletes it.
1618 * This is a helper function for in-band control and fail-open. */
1620 ofproto_delete_flow(struct ofproto *ofproto,
1621 const struct match *target, unsigned int priority)
1625 rule = rule_from_cls_rule(classifier_find_match_exactly(
1626 &ofproto->tables[0].cls, target, priority));
1628 /* No such rule -> success. */
1630 } else if (rule->pending) {
1631 /* An operation on the rule is already pending -> failure.
1632 * Caller must retry later if it's important. */
1635 /* Initiate deletion -> success. */
1636 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1637 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1638 oftable_remove_rule(rule);
1639 ofproto->ofproto_class->rule_destruct(rule);
1640 ofopgroup_submit(group);
1646 /* Starts the process of deleting all of the flows from all of ofproto's flow
1647 * tables and then reintroducing the flows required by in-band control and
1648 * fail-open. The process will complete in a later call to ofproto_run(). */
1650 ofproto_flush_flows(struct ofproto *ofproto)
1652 COVERAGE_INC(ofproto_flush);
1653 ofproto->state = S_FLUSH;
1657 reinit_ports(struct ofproto *p)
1659 struct ofproto_port_dump dump;
1660 struct sset devnames;
1661 struct ofport *ofport;
1662 struct ofproto_port ofproto_port;
1663 const char *devname;
1665 COVERAGE_INC(ofproto_reinit_ports);
1667 sset_init(&devnames);
1668 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1669 sset_add(&devnames, netdev_get_name(ofport->netdev));
1671 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1672 sset_add(&devnames, ofproto_port.name);
1675 SSET_FOR_EACH (devname, &devnames) {
1676 update_port(p, devname);
1678 sset_destroy(&devnames);
1682 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1685 uint16_t end_port_no = ofproto->alloc_port_no;
1687 ofp_port = simap_get(&ofproto->ofp_requests, netdev_name);
1688 ofp_port = ofp_port ? ofp_port : OFPP_NONE;
1690 if (ofp_port >= ofproto->max_ports
1691 || bitmap_is_set(ofproto->ofp_port_ids, ofp_port)) {
1692 /* Search for a free OpenFlow port number. We try not to
1693 * immediately reuse them to prevent problems due to old
1696 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1697 ofproto->alloc_port_no = 0;
1699 if (!bitmap_is_set(ofproto->ofp_port_ids,
1700 ofproto->alloc_port_no)) {
1701 ofp_port = ofproto->alloc_port_no;
1704 if (ofproto->alloc_port_no == end_port_no) {
1709 bitmap_set1(ofproto->ofp_port_ids, ofp_port);
1714 dealloc_ofp_port(const struct ofproto *ofproto, uint16_t ofp_port)
1716 if (ofp_port < ofproto->max_ports) {
1717 bitmap_set0(ofproto->ofp_port_ids, ofp_port);
1721 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1722 * pointer if the netdev cannot be opened. On success, also fills in
1724 static struct netdev *
1725 ofport_open(struct ofproto *ofproto,
1726 struct ofproto_port *ofproto_port,
1727 struct ofputil_phy_port *pp)
1729 enum netdev_flags flags;
1730 struct netdev *netdev;
1733 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1735 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1736 "cannot be opened (%s)",
1738 ofproto_port->name, ofproto_port->ofp_port,
1739 ofproto_port->name, strerror(error));
1743 if (ofproto_port->ofp_port == OFPP_NONE) {
1744 if (!strcmp(ofproto->name, ofproto_port->name)) {
1745 ofproto_port->ofp_port = OFPP_LOCAL;
1747 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1748 ofproto_port->name);
1751 pp->port_no = ofproto_port->ofp_port;
1752 netdev_get_etheraddr(netdev, pp->hw_addr);
1753 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1754 netdev_get_flags(netdev, &flags);
1755 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1756 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1757 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1758 &pp->supported, &pp->peer);
1759 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1760 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1765 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1766 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1769 ofport_equal(const struct ofputil_phy_port *a,
1770 const struct ofputil_phy_port *b)
1772 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1773 && a->state == b->state
1774 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1775 && a->curr == b->curr
1776 && a->advertised == b->advertised
1777 && a->supported == b->supported
1778 && a->peer == b->peer
1779 && a->curr_speed == b->curr_speed
1780 && a->max_speed == b->max_speed);
1783 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1784 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1785 * one with the same name or port number). */
1787 ofport_install(struct ofproto *p,
1788 struct netdev *netdev, const struct ofputil_phy_port *pp)
1790 const char *netdev_name = netdev_get_name(netdev);
1791 struct ofport *ofport;
1794 /* Create ofport. */
1795 ofport = p->ofproto_class->port_alloc();
1800 ofport->ofproto = p;
1801 ofport->netdev = netdev;
1802 ofport->change_seq = netdev_change_seq(netdev);
1804 ofport->ofp_port = pp->port_no;
1805 ofport->created = time_msec();
1807 /* Add port to 'p'. */
1808 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1809 shash_add(&p->port_by_name, netdev_name, ofport);
1811 update_mtu(p, ofport);
1813 /* Let the ofproto_class initialize its private data. */
1814 error = p->ofproto_class->port_construct(ofport);
1818 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1822 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1823 p->name, netdev_name, strerror(error));
1825 ofport_destroy__(ofport);
1827 netdev_close(netdev);
1831 /* Removes 'ofport' from 'p' and destroys it. */
1833 ofport_remove(struct ofport *ofport)
1835 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1837 ofport_destroy(ofport);
1840 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1843 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1845 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1847 ofport_remove(port);
1851 /* Updates 'port' with new 'pp' description.
1853 * Does not handle a name or port number change. The caller must implement
1854 * such a change as a delete followed by an add. */
1856 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1858 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1859 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1860 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1861 port->pp.state = pp->state;
1862 port->pp.curr = pp->curr;
1863 port->pp.advertised = pp->advertised;
1864 port->pp.supported = pp->supported;
1865 port->pp.peer = pp->peer;
1866 port->pp.curr_speed = pp->curr_speed;
1867 port->pp.max_speed = pp->max_speed;
1869 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1872 /* Update OpenFlow 'state' in 'port' and notify controller. */
1874 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1876 if (port->pp.state != state) {
1877 port->pp.state = state;
1878 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1884 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1886 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1888 if (port->ofproto->ofproto_class->set_realdev) {
1889 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1891 if (port->ofproto->ofproto_class->set_stp_port) {
1892 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1894 if (port->ofproto->ofproto_class->set_cfm) {
1895 port->ofproto->ofproto_class->set_cfm(port, NULL);
1897 if (port->ofproto->ofproto_class->bundle_remove) {
1898 port->ofproto->ofproto_class->bundle_remove(port);
1904 ofport_destroy__(struct ofport *port)
1906 struct ofproto *ofproto = port->ofproto;
1907 const char *name = netdev_get_name(port->netdev);
1909 hmap_remove(&ofproto->ports, &port->hmap_node);
1910 shash_delete(&ofproto->port_by_name,
1911 shash_find(&ofproto->port_by_name, name));
1913 netdev_close(port->netdev);
1914 ofproto->ofproto_class->port_dealloc(port);
1918 ofport_destroy(struct ofport *port)
1921 dealloc_ofp_port(port->ofproto, port->ofp_port);
1922 port->ofproto->ofproto_class->port_destruct(port);
1923 ofport_destroy__(port);
1928 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1930 struct ofport *port;
1932 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1933 hash_int(ofp_port, 0), &ofproto->ports) {
1934 if (port->ofp_port == ofp_port) {
1942 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1944 struct ofproto *ofproto = port->ofproto;
1947 if (ofproto->ofproto_class->port_get_stats) {
1948 error = ofproto->ofproto_class->port_get_stats(port, stats);
1957 update_port(struct ofproto *ofproto, const char *name)
1959 struct ofproto_port ofproto_port;
1960 struct ofputil_phy_port pp;
1961 struct netdev *netdev;
1962 struct ofport *port;
1964 COVERAGE_INC(ofproto_update_port);
1966 /* Fetch 'name''s location and properties from the datapath. */
1967 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1968 ? ofport_open(ofproto, &ofproto_port, &pp)
1971 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1972 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1973 struct netdev *old_netdev = port->netdev;
1975 /* 'name' hasn't changed location. Any properties changed? */
1976 if (!ofport_equal(&port->pp, &pp)) {
1977 ofport_modified(port, &pp);
1980 update_mtu(ofproto, port);
1982 /* Install the newly opened netdev in case it has changed.
1983 * Don't close the old netdev yet in case port_modified has to
1984 * remove a retained reference to it.*/
1985 port->netdev = netdev;
1986 port->change_seq = netdev_change_seq(netdev);
1988 if (port->ofproto->ofproto_class->port_modified) {
1989 port->ofproto->ofproto_class->port_modified(port);
1992 netdev_close(old_netdev);
1994 /* If 'port' is nonnull then its name differs from 'name' and thus
1995 * we should delete it. If we think there's a port named 'name'
1996 * then its port number must be wrong now so delete it too. */
1998 ofport_remove(port);
2000 ofport_remove_with_name(ofproto, name);
2001 ofport_install(ofproto, netdev, &pp);
2004 /* Any port named 'name' is gone now. */
2005 ofport_remove_with_name(ofproto, name);
2007 ofproto_port_destroy(&ofproto_port);
2011 init_ports(struct ofproto *p)
2013 struct ofproto_port_dump dump;
2014 struct ofproto_port ofproto_port;
2015 struct shash_node *node, *next;
2017 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2018 const char *name = ofproto_port.name;
2020 if (shash_find(&p->port_by_name, name)) {
2021 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2024 struct ofputil_phy_port pp;
2025 struct netdev *netdev;
2027 /* Check if an OpenFlow port number had been requested. */
2028 node = shash_find(&init_ofp_ports, name);
2030 const struct iface_hint *iface_hint = node->data;
2031 simap_put(&p->ofp_requests, name, iface_hint->ofp_port);
2034 netdev = ofport_open(p, &ofproto_port, &pp);
2036 ofport_install(p, netdev, &pp);
2041 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2042 struct iface_hint *iface_hint = node->data;
2044 if (!strcmp(iface_hint->br_name, p->name)) {
2045 free(iface_hint->br_name);
2046 free(iface_hint->br_type);
2048 shash_delete(&init_ofp_ports, node);
2055 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2056 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2058 find_min_mtu(struct ofproto *p)
2060 struct ofport *ofport;
2063 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2064 struct netdev *netdev = ofport->netdev;
2067 /* Skip any internal ports, since that's what we're trying to
2069 if (!strcmp(netdev_get_type(netdev), "internal")) {
2073 if (netdev_get_mtu(netdev, &dev_mtu)) {
2076 if (!mtu || dev_mtu < mtu) {
2081 return mtu ? mtu: ETH_PAYLOAD_MAX;
2084 /* Update MTU of all datapath devices on 'p' to the minimum of the
2085 * non-datapath ports in event of 'port' added or changed. */
2087 update_mtu(struct ofproto *p, struct ofport *port)
2089 struct ofport *ofport;
2090 struct netdev *netdev = port->netdev;
2091 int dev_mtu, old_min;
2093 if (netdev_get_mtu(netdev, &dev_mtu)) {
2097 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2098 if (dev_mtu > p->min_mtu) {
2099 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2100 dev_mtu = p->min_mtu;
2103 port->mtu = dev_mtu;
2107 /* For non-internal port find new min mtu. */
2108 old_min = p->min_mtu;
2109 port->mtu = dev_mtu;
2110 p->min_mtu = find_min_mtu(p);
2111 if (p->min_mtu == old_min) {
2115 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2116 struct netdev *netdev = ofport->netdev;
2118 if (!strcmp(netdev_get_type(netdev), "internal")) {
2119 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2120 ofport->mtu = p->min_mtu;
2127 ofproto_rule_destroy__(struct rule *rule)
2130 cls_rule_destroy(&rule->cr);
2131 free(rule->ofpacts);
2132 rule->ofproto->ofproto_class->rule_dealloc(rule);
2136 /* This function allows an ofproto implementation to destroy any rules that
2137 * remain when its ->destruct() function is called. The caller must have
2138 * already uninitialized any derived members of 'rule' (step 5 described in the
2139 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2140 * This function implements steps 6 and 7.
2142 * This function should only be called from an ofproto implementation's
2143 * ->destruct() function. It is not suitable elsewhere. */
2145 ofproto_rule_destroy(struct rule *rule)
2147 ovs_assert(!rule->pending);
2148 oftable_remove_rule(rule);
2149 ofproto_rule_destroy__(rule);
2152 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2153 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2155 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
2157 return (port == OFPP_ANY
2158 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2161 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2162 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2164 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
2166 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2171 case OFOPERATION_ADD:
2172 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2174 case OFOPERATION_DELETE:
2177 case OFOPERATION_MODIFY:
2178 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2184 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2185 * statistics appropriately. 'packet' must have at least sizeof(struct
2186 * ofp10_packet_in) bytes of headroom.
2188 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2189 * with statistics for 'packet' either way.
2191 * Takes ownership of 'packet'. */
2193 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
2197 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2199 flow_extract(packet, 0, 0, NULL, in_port, &flow);
2200 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2203 /* Returns true if 'rule' should be hidden from the controller.
2205 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2206 * (e.g. by in-band control) and are intentionally hidden from the
2209 ofproto_rule_is_hidden(const struct rule *rule)
2211 return rule->cr.priority > UINT16_MAX;
2214 static enum oftable_flags
2215 rule_get_flags(const struct rule *rule)
2217 return rule->ofproto->tables[rule->table_id].flags;
2221 rule_is_modifiable(const struct rule *rule)
2223 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2227 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2229 ofconn_send_reply(ofconn, make_echo_reply(oh));
2234 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2236 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2237 struct ofputil_switch_features features;
2238 struct ofport *port;
2244 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2246 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2248 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2249 * if present, are always at the end.) */
2250 n_tables = ofproto->n_tables;
2251 for (i = 0; i < ofproto->n_tables; i++) {
2252 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2258 features.datapath_id = ofproto->datapath_id;
2259 features.n_buffers = pktbuf_capacity();
2260 features.n_tables = n_tables;
2261 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2262 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2264 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2266 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2267 features.auxiliary_id = 0;
2268 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2270 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2271 ofputil_put_switch_features_port(&port->pp, b);
2274 ofconn_send_reply(ofconn, b);
2279 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2281 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2282 struct ofp_switch_config *osc;
2283 enum ofp_config_flags flags;
2287 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2288 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2289 flags = ofproto->frag_handling;
2290 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2291 if (oh->version < OFP13_VERSION
2292 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2293 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2295 osc->flags = htons(flags);
2296 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2297 ofconn_send_reply(ofconn, buf);
2303 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2305 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2306 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2307 uint16_t flags = ntohs(osc->flags);
2309 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2310 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2311 enum ofp_config_flags cur = ofproto->frag_handling;
2312 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2314 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2316 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2317 ofproto->frag_handling = next;
2319 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2321 ofputil_frag_handling_to_string(next));
2325 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2326 ofconn_set_invalid_ttl_to_controller(ofconn,
2327 (oh->version < OFP13_VERSION
2328 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2330 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2335 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2336 * error message code for the caller to propagate upward. Otherwise, returns
2339 * The log message mentions 'msg_type'. */
2341 reject_slave_controller(struct ofconn *ofconn)
2343 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2344 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2345 return OFPERR_OFPBRC_EPERM;
2352 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2354 struct ofproto *p = ofconn_get_ofproto(ofconn);
2355 struct ofputil_packet_out po;
2356 struct ofpbuf *payload;
2357 uint64_t ofpacts_stub[1024 / 8];
2358 struct ofpbuf ofpacts;
2362 COVERAGE_INC(ofproto_packet_out);
2364 error = reject_slave_controller(ofconn);
2369 /* Decode message. */
2370 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2371 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2373 goto exit_free_ofpacts;
2375 if (po.in_port >= p->max_ports && po.in_port < OFPP_MAX) {
2376 error = OFPERR_OFPBRC_BAD_PORT;
2377 goto exit_free_ofpacts;
2382 if (po.buffer_id != UINT32_MAX) {
2383 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2384 if (error || !payload) {
2385 goto exit_free_ofpacts;
2388 payload = xmalloc(sizeof *payload);
2389 ofpbuf_use_const(payload, po.packet, po.packet_len);
2392 /* Verify actions against packet, then send packet if successful. */
2393 flow_extract(payload, 0, 0, NULL, po.in_port, &flow);
2394 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2396 error = p->ofproto_class->packet_out(p, payload, &flow,
2397 po.ofpacts, po.ofpacts_len);
2399 ofpbuf_delete(payload);
2402 ofpbuf_uninit(&ofpacts);
2408 update_port_config(struct ofport *port,
2409 enum ofputil_port_config config,
2410 enum ofputil_port_config mask)
2412 enum ofputil_port_config old_config = port->pp.config;
2413 enum ofputil_port_config toggle;
2415 toggle = (config ^ port->pp.config) & mask;
2416 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2417 if (config & OFPUTIL_PC_PORT_DOWN) {
2418 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2420 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2422 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2425 port->pp.config ^= toggle;
2426 if (port->pp.config != old_config) {
2427 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2432 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2434 struct ofproto *p = ofconn_get_ofproto(ofconn);
2435 struct ofputil_port_mod pm;
2436 struct ofport *port;
2439 error = reject_slave_controller(ofconn);
2444 error = ofputil_decode_port_mod(oh, &pm);
2449 port = ofproto_get_port(p, pm.port_no);
2451 return OFPERR_OFPPMFC_BAD_PORT;
2452 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2453 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2455 update_port_config(port, pm.config, pm.mask);
2457 netdev_set_advertisements(port->netdev, pm.advertise);
2464 handle_desc_stats_request(struct ofconn *ofconn,
2465 const struct ofp_header *request)
2467 static const char *default_mfr_desc = "Nicira, Inc.";
2468 static const char *default_hw_desc = "Open vSwitch";
2469 static const char *default_sw_desc = VERSION;
2470 static const char *default_serial_desc = "None";
2471 static const char *default_dp_desc = "None";
2473 struct ofproto *p = ofconn_get_ofproto(ofconn);
2474 struct ofp_desc_stats *ods;
2477 msg = ofpraw_alloc_stats_reply(request, 0);
2478 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2479 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2480 sizeof ods->mfr_desc);
2481 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2482 sizeof ods->hw_desc);
2483 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2484 sizeof ods->sw_desc);
2485 ovs_strlcpy(ods->serial_num,
2486 p->serial_desc ? p->serial_desc : default_serial_desc,
2487 sizeof ods->serial_num);
2488 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2489 sizeof ods->dp_desc);
2490 ofconn_send_reply(ofconn, msg);
2496 handle_table_stats_request(struct ofconn *ofconn,
2497 const struct ofp_header *request)
2499 struct ofproto *p = ofconn_get_ofproto(ofconn);
2500 struct ofp12_table_stats *ots;
2505 /* Set up default values.
2507 * ofp12_table_stats is used as a generic structure as
2508 * it is able to hold all the fields for ofp10_table_stats
2509 * and ofp11_table_stats (and of course itself).
2511 ots = xcalloc(p->n_tables, sizeof *ots);
2512 for (i = 0; i < p->n_tables; i++) {
2513 ots[i].table_id = i;
2514 sprintf(ots[i].name, "table%zu", i);
2515 ots[i].match = htonll(OFPXMT12_MASK);
2516 ots[i].wildcards = htonll(OFPXMT12_MASK);
2517 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2518 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2519 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2520 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2521 ots[i].metadata_match = htonll(UINT64_MAX);
2522 ots[i].metadata_write = htonll(UINT64_MAX);
2523 ots[i].instructions = htonl(OFPIT11_ALL);
2524 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2525 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2526 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2529 p->ofproto_class->get_tables(p, ots);
2531 /* Post-process the tables, dropping hidden tables. */
2532 n_tables = p->n_tables;
2533 for (i = 0; i < p->n_tables; i++) {
2534 const struct oftable *table = &p->tables[i];
2536 if (table->flags & OFTABLE_HIDDEN) {
2542 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2545 if (table->max_flows < ntohl(ots[i].max_entries)) {
2546 ots[i].max_entries = htonl(table->max_flows);
2550 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2551 ofconn_send_reply(ofconn, msg);
2559 append_port_stat(struct ofport *port, struct list *replies)
2561 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2563 calc_duration(port->created, time_msec(),
2564 &ops.duration_sec, &ops.duration_nsec);
2566 /* Intentionally ignore return value, since errors will set
2567 * 'stats' to all-1s, which is correct for OpenFlow, and
2568 * netdev_get_stats() will log errors. */
2569 ofproto_port_get_stats(port, &ops.stats);
2571 ofputil_append_port_stat(replies, &ops);
2575 handle_port_stats_request(struct ofconn *ofconn,
2576 const struct ofp_header *request)
2578 struct ofproto *p = ofconn_get_ofproto(ofconn);
2579 struct ofport *port;
2580 struct list replies;
2584 error = ofputil_decode_port_stats_request(request, &port_no);
2589 ofpmp_init(&replies, request);
2590 if (port_no != OFPP_ANY) {
2591 port = ofproto_get_port(p, port_no);
2593 append_port_stat(port, &replies);
2596 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2597 append_port_stat(port, &replies);
2601 ofconn_send_replies(ofconn, &replies);
2606 handle_port_desc_stats_request(struct ofconn *ofconn,
2607 const struct ofp_header *request)
2609 struct ofproto *p = ofconn_get_ofproto(ofconn);
2610 enum ofp_version version;
2611 struct ofport *port;
2612 struct list replies;
2614 ofpmp_init(&replies, request);
2616 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2617 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2618 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2621 ofconn_send_replies(ofconn, &replies);
2626 calc_duration(long long int start, long long int now,
2627 uint32_t *sec, uint32_t *nsec)
2629 long long int msecs = now - start;
2630 *sec = msecs / 1000;
2631 *nsec = (msecs % 1000) * (1000 * 1000);
2634 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2635 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2637 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2639 return (table_id == 0xff || table_id < ofproto->n_tables
2641 : OFPERR_OFPBRC_BAD_TABLE_ID);
2645 static struct oftable *
2646 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2648 struct oftable *table;
2650 for (table = &ofproto->tables[table_id];
2651 table < &ofproto->tables[ofproto->n_tables];
2653 if (!(table->flags & OFTABLE_HIDDEN)) {
2661 static struct oftable *
2662 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2664 if (table_id == 0xff) {
2665 return next_visible_table(ofproto, 0);
2666 } else if (table_id < ofproto->n_tables) {
2667 return &ofproto->tables[table_id];
2673 static struct oftable *
2674 next_matching_table(const struct ofproto *ofproto,
2675 const struct oftable *table, uint8_t table_id)
2677 return (table_id == 0xff
2678 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2682 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2684 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2685 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2687 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2688 * only once, for that table. (This can be used to access tables marked
2691 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2692 * entered at all. (Perhaps you should have validated TABLE_ID with
2693 * check_table_id().)
2695 * All parameters are evaluated multiple times.
2697 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2698 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2700 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2702 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2703 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2704 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2707 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2708 * to 'out_port' are included.
2710 * Hidden rules are always omitted.
2712 * Returns 0 on success, otherwise an OpenFlow error code. */
2714 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2715 const struct match *match,
2716 ovs_be64 cookie, ovs_be64 cookie_mask,
2717 uint16_t out_port, struct list *rules)
2719 struct oftable *table;
2723 error = check_table_id(ofproto, table_id);
2729 cls_rule_init(&cr, match, 0);
2730 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2731 struct cls_cursor cursor;
2734 cls_cursor_init(&cursor, &table->cls, &cr);
2735 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2736 if (rule->pending) {
2737 error = OFPROTO_POSTPONE;
2740 if (!ofproto_rule_is_hidden(rule)
2741 && ofproto_rule_has_out_port(rule, out_port)
2742 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2743 list_push_back(rules, &rule->ofproto_node);
2749 cls_rule_destroy(&cr);
2753 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2754 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2755 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2758 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2759 * to 'out_port' are included.
2761 * Hidden rules are always omitted.
2763 * Returns 0 on success, otherwise an OpenFlow error code. */
2765 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2766 const struct match *match, unsigned int priority,
2767 ovs_be64 cookie, ovs_be64 cookie_mask,
2768 uint16_t out_port, struct list *rules)
2770 struct oftable *table;
2774 error = check_table_id(ofproto, table_id);
2780 cls_rule_init(&cr, match, priority);
2781 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2784 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2787 if (rule->pending) {
2788 error = OFPROTO_POSTPONE;
2791 if (!ofproto_rule_is_hidden(rule)
2792 && ofproto_rule_has_out_port(rule, out_port)
2793 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2794 list_push_back(rules, &rule->ofproto_node);
2800 cls_rule_destroy(&cr);
2804 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2805 * forced into the range of a uint16_t. */
2807 age_secs(long long int age_ms)
2809 return (age_ms < 0 ? 0
2810 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2811 : (unsigned int) age_ms / 1000);
2815 handle_flow_stats_request(struct ofconn *ofconn,
2816 const struct ofp_header *request)
2818 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2819 struct ofputil_flow_stats_request fsr;
2820 struct list replies;
2825 error = ofputil_decode_flow_stats_request(&fsr, request);
2830 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2831 fsr.cookie, fsr.cookie_mask,
2832 fsr.out_port, &rules);
2837 ofpmp_init(&replies, request);
2838 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2839 long long int now = time_msec();
2840 struct ofputil_flow_stats fs;
2842 minimatch_expand(&rule->cr.match, &fs.match);
2843 fs.priority = rule->cr.priority;
2844 fs.cookie = rule->flow_cookie;
2845 fs.table_id = rule->table_id;
2846 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
2847 fs.idle_timeout = rule->idle_timeout;
2848 fs.hard_timeout = rule->hard_timeout;
2849 fs.idle_age = age_secs(now - rule->used);
2850 fs.hard_age = age_secs(now - rule->modified);
2851 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2853 fs.ofpacts = rule->ofpacts;
2854 fs.ofpacts_len = rule->ofpacts_len;
2856 if (rule->send_flow_removed) {
2857 fs.flags |= OFPFF_SEND_FLOW_REM;
2858 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2859 and OFPFF13_NO_BYT_COUNTS */
2861 ofputil_append_flow_stats_reply(&fs, &replies);
2863 ofconn_send_replies(ofconn, &replies);
2869 flow_stats_ds(struct rule *rule, struct ds *results)
2871 uint64_t packet_count, byte_count;
2873 rule->ofproto->ofproto_class->rule_get_stats(rule,
2874 &packet_count, &byte_count);
2876 if (rule->table_id != 0) {
2877 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2879 ds_put_format(results, "duration=%llds, ",
2880 (time_msec() - rule->created) / 1000);
2881 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2882 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2883 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2884 cls_rule_format(&rule->cr, results);
2885 ds_put_char(results, ',');
2886 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2887 ds_put_cstr(results, "\n");
2890 /* Adds a pretty-printed description of all flows to 'results', including
2891 * hidden flows (e.g., set up by in-band control). */
2893 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2895 struct oftable *table;
2897 OFPROTO_FOR_EACH_TABLE (table, p) {
2898 struct cls_cursor cursor;
2901 cls_cursor_init(&cursor, &table->cls, NULL);
2902 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2903 flow_stats_ds(rule, results);
2908 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2909 * '*engine_type' and '*engine_id', respectively. */
2911 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2912 uint8_t *engine_type, uint8_t *engine_id)
2914 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2917 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
2918 * true if the port's CFM status was successfully stored into '*status'.
2919 * Returns false if the port did not have CFM configured, in which case
2920 * '*status' is indeterminate.
2922 * The caller must provide and owns '*status', but it does not own and must not
2923 * modify or free the array returned in 'status->rmps'. */
2925 ofproto_port_get_cfm_status(const struct ofproto *ofproto, uint16_t ofp_port,
2926 struct ofproto_cfm_status *status)
2928 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2930 && ofproto->ofproto_class->get_cfm_status
2931 && ofproto->ofproto_class->get_cfm_status(ofport, status));
2935 handle_aggregate_stats_request(struct ofconn *ofconn,
2936 const struct ofp_header *oh)
2938 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2939 struct ofputil_flow_stats_request request;
2940 struct ofputil_aggregate_stats stats;
2941 bool unknown_packets, unknown_bytes;
2942 struct ofpbuf *reply;
2947 error = ofputil_decode_flow_stats_request(&request, oh);
2952 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2953 request.cookie, request.cookie_mask,
2954 request.out_port, &rules);
2959 memset(&stats, 0, sizeof stats);
2960 unknown_packets = unknown_bytes = false;
2961 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2962 uint64_t packet_count;
2963 uint64_t byte_count;
2965 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2968 if (packet_count == UINT64_MAX) {
2969 unknown_packets = true;
2971 stats.packet_count += packet_count;
2974 if (byte_count == UINT64_MAX) {
2975 unknown_bytes = true;
2977 stats.byte_count += byte_count;
2982 if (unknown_packets) {
2983 stats.packet_count = UINT64_MAX;
2985 if (unknown_bytes) {
2986 stats.byte_count = UINT64_MAX;
2989 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2990 ofconn_send_reply(ofconn, reply);
2995 struct queue_stats_cbdata {
2996 struct ofport *ofport;
2997 struct list replies;
3001 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3002 const struct netdev_queue_stats *stats)
3005 struct ofputil_queue_stats oqs = {
3006 .port_no = cbdata->ofport->pp.port_no,
3007 .queue_id = queue_id,
3010 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3014 handle_queue_stats_dump_cb(uint32_t queue_id,
3015 struct netdev_queue_stats *stats,
3018 struct queue_stats_cbdata *cbdata = cbdata_;
3020 put_queue_stats(cbdata, queue_id, stats);
3024 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3025 struct queue_stats_cbdata *cbdata)
3027 cbdata->ofport = port;
3028 if (queue_id == OFPQ_ALL) {
3029 netdev_dump_queue_stats(port->netdev,
3030 handle_queue_stats_dump_cb, cbdata);
3032 struct netdev_queue_stats stats;
3034 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3035 put_queue_stats(cbdata, queue_id, &stats);
3037 return OFPERR_OFPQOFC_BAD_QUEUE;
3044 handle_queue_stats_request(struct ofconn *ofconn,
3045 const struct ofp_header *rq)
3047 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3048 struct queue_stats_cbdata cbdata;
3049 struct ofport *port;
3051 struct ofputil_queue_stats_request oqsr;
3053 COVERAGE_INC(ofproto_queue_req);
3055 ofpmp_init(&cbdata.replies, rq);
3057 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3062 if (oqsr.port_no == OFPP_ANY) {
3063 error = OFPERR_OFPQOFC_BAD_QUEUE;
3064 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3065 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3070 port = ofproto_get_port(ofproto, oqsr.port_no);
3072 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3073 : OFPERR_OFPQOFC_BAD_PORT);
3076 ofconn_send_replies(ofconn, &cbdata.replies);
3078 ofpbuf_list_delete(&cbdata.replies);
3085 is_flow_deletion_pending(const struct ofproto *ofproto,
3086 const struct cls_rule *cls_rule,
3089 if (!hmap_is_empty(&ofproto->deletions)) {
3090 struct ofoperation *op;
3092 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3093 cls_rule_hash(cls_rule, table_id),
3094 &ofproto->deletions) {
3095 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3104 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3105 * in which no matching flow already exists in the flow table.
3107 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3108 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3109 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3110 * initiated now but may be retried later.
3112 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3113 * ownership remains with the caller.
3115 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3118 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3119 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3121 struct oftable *table;
3122 struct ofopgroup *group;
3123 struct rule *victim;
3128 error = check_table_id(ofproto, fm->table_id);
3134 if (fm->table_id == 0xff) {
3135 if (ofproto->ofproto_class->rule_choose_table) {
3136 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3142 ovs_assert(table_id < ofproto->n_tables);
3146 } else if (fm->table_id < ofproto->n_tables) {
3147 table_id = fm->table_id;
3149 return OFPERR_OFPBRC_BAD_TABLE_ID;
3152 table = &ofproto->tables[table_id];
3154 if (table->flags & OFTABLE_READONLY) {
3155 return OFPERR_OFPBRC_EPERM;
3158 /* Allocate new rule and initialize classifier rule. */
3159 rule = ofproto->ofproto_class->rule_alloc();
3161 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3162 ofproto->name, strerror(error));
3165 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3167 /* Serialize against pending deletion. */
3168 if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) {
3169 cls_rule_destroy(&rule->cr);
3170 ofproto->ofproto_class->rule_dealloc(rule);
3171 return OFPROTO_POSTPONE;
3174 /* Check for overlap, if requested. */
3175 if (fm->flags & OFPFF_CHECK_OVERLAP
3176 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3177 cls_rule_destroy(&rule->cr);
3178 ofproto->ofproto_class->rule_dealloc(rule);
3179 return OFPERR_OFPFMFC_OVERLAP;
3182 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3184 rule->ofproto = ofproto;
3185 rule->pending = NULL;
3186 rule->flow_cookie = fm->new_cookie;
3187 rule->created = rule->modified = rule->used = time_msec();
3188 rule->idle_timeout = fm->idle_timeout;
3189 rule->hard_timeout = fm->hard_timeout;
3190 rule->table_id = table - ofproto->tables;
3191 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3192 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3193 and OFPFF13_NO_BYT_COUNTS */
3194 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3195 rule->ofpacts_len = fm->ofpacts_len;
3196 rule->evictable = true;
3197 rule->eviction_group = NULL;
3198 list_init(&rule->expirable);
3199 rule->monitor_flags = 0;
3200 rule->add_seqno = 0;
3201 rule->modify_seqno = 0;
3203 /* Insert new rule. */
3204 victim = oftable_replace_rule(rule);
3205 if (victim && !rule_is_modifiable(victim)) {
3206 error = OFPERR_OFPBRC_EPERM;
3207 } else if (victim && victim->pending) {
3208 error = OFPROTO_POSTPONE;
3210 struct ofoperation *op;
3213 if (classifier_count(&table->cls) > table->max_flows) {
3216 was_evictable = rule->evictable;
3217 rule->evictable = false;
3218 evict = choose_rule_to_evict(table);
3219 rule->evictable = was_evictable;
3222 error = OFPERR_OFPFMFC_TABLE_FULL;
3224 } else if (evict->pending) {
3225 error = OFPROTO_POSTPONE;
3232 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3233 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3234 op->victim = victim;
3236 error = ofproto->ofproto_class->rule_construct(rule);
3238 op->group->n_running--;
3239 ofoperation_destroy(rule->pending);
3241 delete_flow__(evict, group);
3243 ofopgroup_submit(group);
3247 /* Back out if an error occurred. */
3249 oftable_substitute_rule(rule, victim);
3250 ofproto_rule_destroy__(rule);
3255 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3257 /* Modifies the rules listed in 'rules', changing their actions to match those
3260 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3263 * Returns 0 on success, otherwise an OpenFlow error code. */
3265 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3266 const struct ofputil_flow_mod *fm,
3267 const struct ofp_header *request, struct list *rules)
3269 struct ofopgroup *group;
3273 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3274 error = OFPERR_OFPBRC_EPERM;
3275 LIST_FOR_EACH (rule, ofproto_node, rules) {
3276 struct ofoperation *op;
3277 bool actions_changed;
3278 ovs_be64 new_cookie;
3280 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3282 if (rule_is_modifiable(rule)) {
3283 /* At least one rule is modifiable, don't report EPERM error. */
3289 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3290 rule->ofpacts, rule->ofpacts_len);
3291 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
3293 : rule->flow_cookie);
3295 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3296 rule->flow_cookie = new_cookie;
3297 if (actions_changed) {
3298 op->ofpacts = rule->ofpacts;
3299 op->ofpacts_len = rule->ofpacts_len;
3300 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3301 rule->ofpacts_len = fm->ofpacts_len;
3302 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3304 ofoperation_complete(op, 0);
3307 ofopgroup_submit(group);
3313 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3314 const struct ofputil_flow_mod *fm,
3315 const struct ofp_header *request)
3317 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3320 return add_flow(ofproto, ofconn, fm, request);
3323 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3326 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3329 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3330 const struct ofputil_flow_mod *fm,
3331 const struct ofp_header *request)
3336 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3337 fm->cookie, fm->cookie_mask,
3341 } else if (list_is_empty(&rules)) {
3342 return modify_flows_add(ofproto, ofconn, fm, request);
3344 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3348 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3351 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3354 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3355 const struct ofputil_flow_mod *fm,
3356 const struct ofp_header *request)
3361 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3362 fm->priority, fm->cookie, fm->cookie_mask,
3367 } else if (list_is_empty(&rules)) {
3368 return modify_flows_add(ofproto, ofconn, fm, request);
3370 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3371 fm, request, &rules)
3376 /* OFPFC_DELETE implementation. */
3379 delete_flow__(struct rule *rule, struct ofopgroup *group)
3381 struct ofproto *ofproto = rule->ofproto;
3383 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3385 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3386 oftable_remove_rule(rule);
3387 ofproto->ofproto_class->rule_destruct(rule);
3390 /* Deletes the rules listed in 'rules'.
3392 * Returns 0 on success, otherwise an OpenFlow error code. */
3394 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3395 const struct ofp_header *request, struct list *rules)
3397 struct rule *rule, *next;
3398 struct ofopgroup *group;
3400 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3401 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3402 delete_flow__(rule, group);
3404 ofopgroup_submit(group);
3409 /* Implements OFPFC_DELETE. */
3411 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3412 const struct ofputil_flow_mod *fm,
3413 const struct ofp_header *request)
3418 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3419 fm->cookie, fm->cookie_mask,
3420 fm->out_port, &rules);
3421 return (error ? error
3422 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3427 /* Implements OFPFC_DELETE_STRICT. */
3429 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3430 const struct ofputil_flow_mod *fm,
3431 const struct ofp_header *request)
3436 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3437 fm->priority, fm->cookie, fm->cookie_mask,
3438 fm->out_port, &rules);
3439 return (error ? error
3440 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3446 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3448 struct ofputil_flow_removed fr;
3450 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3454 minimatch_expand(&rule->cr.match, &fr.match);
3455 fr.priority = rule->cr.priority;
3456 fr.cookie = rule->flow_cookie;
3458 fr.table_id = rule->table_id;
3459 calc_duration(rule->created, time_msec(),
3460 &fr.duration_sec, &fr.duration_nsec);
3461 fr.idle_timeout = rule->idle_timeout;
3462 fr.hard_timeout = rule->hard_timeout;
3463 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3466 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3470 ofproto_rule_update_used(struct rule *rule, long long int used)
3472 if (used > rule->used) {
3473 struct eviction_group *evg = rule->eviction_group;
3477 heap_change(&evg->rules, &rule->evg_node,
3478 rule_eviction_priority(rule));
3483 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3484 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3487 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3490 * ofproto implementation ->run() functions should use this function to expire
3491 * OpenFlow flows. */
3493 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3495 struct ofproto *ofproto = rule->ofproto;
3496 struct ofopgroup *group;
3498 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3500 ofproto_rule_send_removed(rule, reason);
3502 group = ofopgroup_create_unattached(ofproto);
3503 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3504 oftable_remove_rule(rule);
3505 ofproto->ofproto_class->rule_destruct(rule);
3506 ofopgroup_submit(group);
3510 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3512 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3513 struct ofputil_flow_mod fm;
3514 uint64_t ofpacts_stub[1024 / 8];
3515 struct ofpbuf ofpacts;
3519 error = reject_slave_controller(ofconn);
3524 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3525 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3528 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3529 &fm.match.flow, ofproto->max_ports);
3532 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3535 goto exit_free_ofpacts;
3538 /* Record the operation for logging a summary report. */
3539 switch (fm.command) {
3545 case OFPFC_MODIFY_STRICT:
3546 ofproto->n_modify++;
3550 case OFPFC_DELETE_STRICT:
3551 ofproto->n_delete++;
3556 if (ofproto->next_op_report == LLONG_MAX) {
3557 ofproto->first_op = now;
3558 ofproto->next_op_report = MAX(now + 10 * 1000,
3559 ofproto->op_backoff);
3560 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3562 ofproto->last_op = now;
3565 ofpbuf_uninit(&ofpacts);
3571 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3572 const struct ofputil_flow_mod *fm,
3573 const struct ofp_header *oh)
3575 if (ofproto->n_pending >= 50) {
3576 ovs_assert(!list_is_empty(&ofproto->pending));
3577 return OFPROTO_POSTPONE;
3580 switch (fm->command) {
3582 return add_flow(ofproto, ofconn, fm, oh);
3585 return modify_flows_loose(ofproto, ofconn, fm, oh);
3587 case OFPFC_MODIFY_STRICT:
3588 return modify_flow_strict(ofproto, ofconn, fm, oh);
3591 return delete_flows_loose(ofproto, ofconn, fm, oh);
3593 case OFPFC_DELETE_STRICT:
3594 return delete_flow_strict(ofproto, ofconn, fm, oh);
3597 if (fm->command > 0xff) {
3598 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3599 "flow_mod_table_id extension is not enabled",
3602 return OFPERR_OFPFMFC_BAD_COMMAND;
3607 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3609 struct ofputil_role_request request;
3610 struct ofputil_role_request reply;
3614 error = ofputil_decode_role_message(oh, &request);
3619 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3620 if (ofconn_get_role(ofconn) != request.role
3621 && ofconn_has_pending_opgroups(ofconn)) {
3622 return OFPROTO_POSTPONE;
3625 if (request.have_generation_id
3626 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3627 return OFPERR_OFPRRFC_STALE;
3630 ofconn_set_role(ofconn, request.role);
3633 reply.role = ofconn_get_role(ofconn);
3634 reply.have_generation_id = ofconn_get_master_election_id(
3635 ofconn, &reply.generation_id);
3636 buf = ofputil_encode_role_reply(oh, &reply);
3637 ofconn_send_reply(ofconn, buf);
3643 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3644 const struct ofp_header *oh)
3646 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3647 enum ofputil_protocol cur, next;
3649 cur = ofconn_get_protocol(ofconn);
3650 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3651 ofconn_set_protocol(ofconn, next);
3657 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3659 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3660 enum ofputil_protocol cur, next;
3661 enum ofputil_protocol next_base;
3663 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3665 return OFPERR_OFPBRC_EPERM;
3668 cur = ofconn_get_protocol(ofconn);
3669 next = ofputil_protocol_set_base(cur, next_base);
3670 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3671 /* Avoid sending async messages in surprising protocol. */
3672 return OFPROTO_POSTPONE;
3675 ofconn_set_protocol(ofconn, next);
3680 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3681 const struct ofp_header *oh)
3683 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3686 format = ntohl(msg->format);
3687 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3688 return OFPERR_OFPBRC_EPERM;
3691 if (format != ofconn_get_packet_in_format(ofconn)
3692 && ofconn_has_pending_opgroups(ofconn)) {
3693 /* Avoid sending async message in surprsing packet in format. */
3694 return OFPROTO_POSTPONE;
3697 ofconn_set_packet_in_format(ofconn, format);
3702 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3704 const struct nx_async_config *msg = ofpmsg_body(oh);
3705 uint32_t master[OAM_N_TYPES];
3706 uint32_t slave[OAM_N_TYPES];
3708 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3709 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3710 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3712 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3713 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3714 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3716 ofconn_set_async_config(ofconn, master, slave);
3717 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3718 !ofconn_get_miss_send_len(ofconn)) {
3719 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3726 handle_nxt_set_controller_id(struct ofconn *ofconn,
3727 const struct ofp_header *oh)
3729 const struct nx_controller_id *nci = ofpmsg_body(oh);
3731 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3732 return OFPERR_NXBRC_MUST_BE_ZERO;
3735 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3740 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3744 if (ofconn_has_pending_opgroups(ofconn)) {
3745 return OFPROTO_POSTPONE;
3748 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3749 ? OFPRAW_OFPT10_BARRIER_REPLY
3750 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3751 ofconn_send_reply(ofconn, buf);
3756 ofproto_compose_flow_refresh_update(const struct rule *rule,
3757 enum nx_flow_monitor_flags flags,
3760 struct ofoperation *op = rule->pending;
3761 struct ofputil_flow_update fu;
3764 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3765 /* We'll report the final flow when the operation completes. Reporting
3766 * it now would cause a duplicate report later. */
3770 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3771 ? NXFME_ADDED : NXFME_MODIFIED);
3773 fu.idle_timeout = rule->idle_timeout;
3774 fu.hard_timeout = rule->hard_timeout;
3775 fu.table_id = rule->table_id;
3776 fu.cookie = rule->flow_cookie;
3777 minimatch_expand(&rule->cr.match, &match);
3779 fu.priority = rule->cr.priority;
3780 if (!(flags & NXFMF_ACTIONS)) {
3784 fu.ofpacts = rule->ofpacts;
3785 fu.ofpacts_len = rule->ofpacts_len;
3787 /* An operation is in progress. Use the previous version of the flow's
3788 * actions, so that when the operation commits we report the change. */
3790 case OFOPERATION_ADD:
3791 /* We already verified that there was a victim. */
3792 fu.ofpacts = op->victim->ofpacts;
3793 fu.ofpacts_len = op->victim->ofpacts_len;
3796 case OFOPERATION_MODIFY:
3798 fu.ofpacts = op->ofpacts;
3799 fu.ofpacts_len = op->ofpacts_len;
3801 fu.ofpacts = rule->ofpacts;
3802 fu.ofpacts_len = rule->ofpacts_len;
3806 case OFOPERATION_DELETE:
3807 fu.ofpacts = rule->ofpacts;
3808 fu.ofpacts_len = rule->ofpacts_len;
3816 if (list_is_empty(msgs)) {
3817 ofputil_start_flow_update(msgs);
3819 ofputil_append_flow_update(&fu, msgs);
3823 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3827 LIST_FOR_EACH (rule, ofproto_node, rules) {
3828 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3829 rule->monitor_flags = 0;
3831 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3836 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3837 struct rule *rule, uint64_t seqno,
3840 enum nx_flow_monitor_flags update;
3842 if (ofproto_rule_is_hidden(rule)) {
3847 ? ofoperation_has_out_port(rule->pending, m->out_port)
3848 : ofproto_rule_has_out_port(rule, m->out_port))) {
3853 if (rule->add_seqno > seqno) {
3854 update = NXFMF_ADD | NXFMF_MODIFY;
3855 } else if (rule->modify_seqno > seqno) {
3856 update = NXFMF_MODIFY;
3861 if (!(m->flags & update)) {
3865 update = NXFMF_INITIAL;
3868 if (!rule->monitor_flags) {
3869 list_push_back(rules, &rule->ofproto_node);
3871 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3875 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3879 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3880 const struct ofoperation *op;
3881 const struct oftable *table;
3882 struct cls_rule target;
3884 cls_rule_init_from_minimatch(&target, &m->match, 0);
3885 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3886 struct cls_cursor cursor;
3889 cls_cursor_init(&cursor, &table->cls, &target);
3890 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3891 ovs_assert(!rule->pending); /* XXX */
3892 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3896 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3897 struct rule *rule = op->rule;
3899 if (((m->table_id == 0xff
3900 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3901 : m->table_id == rule->table_id))
3902 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
3903 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3906 cls_rule_destroy(&target);
3910 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3913 if (m->flags & NXFMF_INITIAL) {
3914 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3919 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3920 uint64_t seqno, struct list *rules)
3922 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3926 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3928 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3929 struct ofmonitor **monitors;
3930 size_t n_monitors, allocated_monitors;
3931 struct list replies;
3938 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3940 n_monitors = allocated_monitors = 0;
3942 struct ofputil_flow_monitor_request request;
3943 struct ofmonitor *m;
3946 retval = ofputil_decode_flow_monitor_request(&request, &b);
3947 if (retval == EOF) {
3949 } else if (retval) {
3954 if (request.table_id != 0xff
3955 && request.table_id >= ofproto->n_tables) {
3956 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3960 error = ofmonitor_create(&request, ofconn, &m);
3965 if (n_monitors >= allocated_monitors) {
3966 monitors = x2nrealloc(monitors, &allocated_monitors,
3969 monitors[n_monitors++] = m;
3973 for (i = 0; i < n_monitors; i++) {
3974 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3977 ofpmp_init(&replies, oh);
3978 ofmonitor_compose_refresh_updates(&rules, &replies);
3979 ofconn_send_replies(ofconn, &replies);
3986 for (i = 0; i < n_monitors; i++) {
3987 ofmonitor_destroy(monitors[i]);
3994 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3996 struct ofmonitor *m;
3999 id = ofputil_decode_flow_monitor_cancel(oh);
4000 m = ofmonitor_lookup(ofconn, id);
4002 return OFPERR_NXBRC_FM_BAD_ID;
4005 ofmonitor_destroy(m);
4010 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4012 const struct ofp_header *oh = msg->data;
4016 error = ofptype_decode(&type, oh);
4022 /* OpenFlow requests. */
4023 case OFPTYPE_ECHO_REQUEST:
4024 return handle_echo_request(ofconn, oh);
4026 case OFPTYPE_FEATURES_REQUEST:
4027 return handle_features_request(ofconn, oh);
4029 case OFPTYPE_GET_CONFIG_REQUEST:
4030 return handle_get_config_request(ofconn, oh);
4032 case OFPTYPE_SET_CONFIG:
4033 return handle_set_config(ofconn, oh);
4035 case OFPTYPE_PACKET_OUT:
4036 return handle_packet_out(ofconn, oh);
4038 case OFPTYPE_PORT_MOD:
4039 return handle_port_mod(ofconn, oh);
4041 case OFPTYPE_FLOW_MOD:
4042 return handle_flow_mod(ofconn, oh);
4044 case OFPTYPE_BARRIER_REQUEST:
4045 return handle_barrier_request(ofconn, oh);
4047 case OFPTYPE_ROLE_REQUEST:
4048 return handle_role_request(ofconn, oh);
4050 /* OpenFlow replies. */
4051 case OFPTYPE_ECHO_REPLY:
4054 /* Nicira extension requests. */
4055 case OFPTYPE_FLOW_MOD_TABLE_ID:
4056 return handle_nxt_flow_mod_table_id(ofconn, oh);
4058 case OFPTYPE_SET_FLOW_FORMAT:
4059 return handle_nxt_set_flow_format(ofconn, oh);
4061 case OFPTYPE_SET_PACKET_IN_FORMAT:
4062 return handle_nxt_set_packet_in_format(ofconn, oh);
4064 case OFPTYPE_SET_CONTROLLER_ID:
4065 return handle_nxt_set_controller_id(ofconn, oh);
4067 case OFPTYPE_FLOW_AGE:
4068 /* Nothing to do. */
4071 case OFPTYPE_FLOW_MONITOR_CANCEL:
4072 return handle_flow_monitor_cancel(ofconn, oh);
4074 case OFPTYPE_SET_ASYNC_CONFIG:
4075 return handle_nxt_set_async_config(ofconn, oh);
4077 /* Statistics requests. */
4078 case OFPTYPE_DESC_STATS_REQUEST:
4079 return handle_desc_stats_request(ofconn, oh);
4081 case OFPTYPE_FLOW_STATS_REQUEST:
4082 return handle_flow_stats_request(ofconn, oh);
4084 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4085 return handle_aggregate_stats_request(ofconn, oh);
4087 case OFPTYPE_TABLE_STATS_REQUEST:
4088 return handle_table_stats_request(ofconn, oh);
4090 case OFPTYPE_PORT_STATS_REQUEST:
4091 return handle_port_stats_request(ofconn, oh);
4093 case OFPTYPE_QUEUE_STATS_REQUEST:
4094 return handle_queue_stats_request(ofconn, oh);
4096 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4097 return handle_port_desc_stats_request(ofconn, oh);
4099 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4100 return handle_flow_monitor_request(ofconn, oh);
4102 /* FIXME: Change the following once they are implemented: */
4103 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4104 case OFPTYPE_GET_ASYNC_REQUEST:
4105 case OFPTYPE_METER_MOD:
4106 case OFPTYPE_GROUP_REQUEST:
4107 case OFPTYPE_GROUP_DESC_REQUEST:
4108 case OFPTYPE_GROUP_FEATURES_REQUEST:
4109 case OFPTYPE_METER_REQUEST:
4110 case OFPTYPE_METER_CONFIG_REQUEST:
4111 case OFPTYPE_METER_FEATURES_REQUEST:
4112 case OFPTYPE_TABLE_FEATURES_REQUEST:
4113 return OFPERR_OFPBRC_BAD_TYPE;
4117 case OFPTYPE_FEATURES_REPLY:
4118 case OFPTYPE_GET_CONFIG_REPLY:
4119 case OFPTYPE_PACKET_IN:
4120 case OFPTYPE_FLOW_REMOVED:
4121 case OFPTYPE_PORT_STATUS:
4122 case OFPTYPE_BARRIER_REPLY:
4123 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4124 case OFPTYPE_DESC_STATS_REPLY:
4125 case OFPTYPE_FLOW_STATS_REPLY:
4126 case OFPTYPE_QUEUE_STATS_REPLY:
4127 case OFPTYPE_PORT_STATS_REPLY:
4128 case OFPTYPE_TABLE_STATS_REPLY:
4129 case OFPTYPE_AGGREGATE_STATS_REPLY:
4130 case OFPTYPE_PORT_DESC_STATS_REPLY:
4131 case OFPTYPE_ROLE_REPLY:
4132 case OFPTYPE_FLOW_MONITOR_PAUSED:
4133 case OFPTYPE_FLOW_MONITOR_RESUMED:
4134 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4135 case OFPTYPE_GET_ASYNC_REPLY:
4136 case OFPTYPE_GROUP_REPLY:
4137 case OFPTYPE_GROUP_DESC_REPLY:
4138 case OFPTYPE_GROUP_FEATURES_REPLY:
4139 case OFPTYPE_METER_REPLY:
4140 case OFPTYPE_METER_CONFIG_REPLY:
4141 case OFPTYPE_METER_FEATURES_REPLY:
4142 case OFPTYPE_TABLE_FEATURES_REPLY:
4144 return OFPERR_OFPBRC_BAD_TYPE;
4149 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4151 int error = handle_openflow__(ofconn, ofp_msg);
4152 if (error && error != OFPROTO_POSTPONE) {
4153 ofconn_send_error(ofconn, ofp_msg->data, error);
4155 COVERAGE_INC(ofproto_recv_openflow);
4156 return error != OFPROTO_POSTPONE;
4159 /* Asynchronous operations. */
4161 /* Creates and returns a new ofopgroup that is not associated with any
4162 * OpenFlow connection.
4164 * The caller should add operations to the returned group with
4165 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4166 static struct ofopgroup *
4167 ofopgroup_create_unattached(struct ofproto *ofproto)
4169 struct ofopgroup *group = xzalloc(sizeof *group);
4170 group->ofproto = ofproto;
4171 list_init(&group->ofproto_node);
4172 list_init(&group->ops);
4173 list_init(&group->ofconn_node);
4177 /* Creates and returns a new ofopgroup for 'ofproto'.
4179 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4180 * connection. The 'request' and 'buffer_id' arguments are ignored.
4182 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4183 * If the ofopgroup eventually fails, then the error reply will include
4184 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4185 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4187 * The caller should add operations to the returned group with
4188 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4189 static struct ofopgroup *
4190 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4191 const struct ofp_header *request, uint32_t buffer_id)
4193 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4195 size_t request_len = ntohs(request->length);
4197 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4199 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4200 group->ofconn = ofconn;
4201 group->request = xmemdup(request, MIN(request_len, 64));
4202 group->buffer_id = buffer_id;
4207 /* Submits 'group' for processing.
4209 * If 'group' contains no operations (e.g. none were ever added, or all of the
4210 * ones that were added completed synchronously), then it is destroyed
4211 * immediately. Otherwise it is added to the ofproto's list of pending
4214 ofopgroup_submit(struct ofopgroup *group)
4216 if (!group->n_running) {
4217 ofopgroup_complete(group);
4219 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4220 group->ofproto->n_pending++;
4225 ofopgroup_complete(struct ofopgroup *group)
4227 struct ofproto *ofproto = group->ofproto;
4229 struct ofconn *abbrev_ofconn;
4230 ovs_be32 abbrev_xid;
4232 struct ofoperation *op, *next_op;
4235 ovs_assert(!group->n_running);
4238 LIST_FOR_EACH (op, group_node, &group->ops) {
4245 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4246 LIST_FOR_EACH (op, group_node, &group->ops) {
4247 if (op->type != OFOPERATION_DELETE) {
4248 struct ofpbuf *packet;
4251 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4255 error = rule_execute(op->rule, in_port, packet);
4262 if (!error && !list_is_empty(&group->ofconn_node)) {
4263 abbrev_ofconn = group->ofconn;
4264 abbrev_xid = group->request->xid;
4266 abbrev_ofconn = NULL;
4267 abbrev_xid = htonl(0);
4269 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4270 struct rule *rule = op->rule;
4272 /* We generally want to report the change to active OpenFlow flow
4273 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4275 - The operation failed.
4277 - The affected rule is not visible to controllers.
4279 - The operation's only effect was to update rule->modified. */
4281 || ofproto_rule_is_hidden(rule)
4282 || (op->type == OFOPERATION_MODIFY
4284 && rule->flow_cookie == op->flow_cookie))) {
4285 /* Check that we can just cast from ofoperation_type to
4286 * nx_flow_update_event. */
4287 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4289 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4291 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4294 ofmonitor_report(ofproto->connmgr, rule,
4295 (enum nx_flow_update_event) op->type,
4296 op->reason, abbrev_ofconn, abbrev_xid);
4299 rule->pending = NULL;
4302 case OFOPERATION_ADD:
4306 ofproto_rule_destroy__(op->victim);
4307 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4308 if (vid_mask == VLAN_VID_MASK) {
4309 if (ofproto->vlan_bitmap) {
4310 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4311 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4312 bitmap_set1(ofproto->vlan_bitmap, vid);
4313 ofproto->vlans_changed = true;
4316 ofproto->vlans_changed = true;
4320 oftable_substitute_rule(rule, op->victim);
4321 ofproto_rule_destroy__(rule);
4325 case OFOPERATION_DELETE:
4326 ovs_assert(!op->error);
4327 ofproto_rule_destroy__(rule);
4331 case OFOPERATION_MODIFY:
4333 rule->modified = time_msec();
4335 rule->flow_cookie = op->flow_cookie;
4337 free(rule->ofpacts);
4338 rule->ofpacts = op->ofpacts;
4339 rule->ofpacts_len = op->ofpacts_len;
4341 op->ofpacts_len = 0;
4350 ofoperation_destroy(op);
4353 ofmonitor_flush(ofproto->connmgr);
4355 if (!list_is_empty(&group->ofproto_node)) {
4356 ovs_assert(ofproto->n_pending > 0);
4357 ofproto->n_pending--;
4358 list_remove(&group->ofproto_node);
4360 if (!list_is_empty(&group->ofconn_node)) {
4361 list_remove(&group->ofconn_node);
4363 ofconn_send_error(group->ofconn, group->request, error);
4365 connmgr_retry(ofproto->connmgr);
4367 free(group->request);
4371 /* Initiates a new operation on 'rule', of the specified 'type', within
4372 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4374 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4375 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4377 * Returns the newly created ofoperation (which is also available as
4378 * rule->pending). */
4379 static struct ofoperation *
4380 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4381 enum ofoperation_type type,
4382 enum ofp_flow_removed_reason reason)
4384 struct ofproto *ofproto = group->ofproto;
4385 struct ofoperation *op;
4387 ovs_assert(!rule->pending);
4389 op = rule->pending = xzalloc(sizeof *op);
4391 list_push_back(&group->ops, &op->group_node);
4394 op->reason = reason;
4395 op->flow_cookie = rule->flow_cookie;
4399 if (type == OFOPERATION_DELETE) {
4400 hmap_insert(&ofproto->deletions, &op->hmap_node,
4401 cls_rule_hash(&rule->cr, rule->table_id));
4408 ofoperation_destroy(struct ofoperation *op)
4410 struct ofopgroup *group = op->group;
4413 op->rule->pending = NULL;
4415 if (op->type == OFOPERATION_DELETE) {
4416 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4418 list_remove(&op->group_node);
4423 /* Indicates that 'op' completed with status 'error', which is either 0 to
4424 * indicate success or an OpenFlow error code on failure.
4426 * If 'error' is 0, indicating success, the operation will be committed
4427 * permanently to the flow table. There is one interesting subcase:
4429 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4430 * the flow table (the "victim" rule) by a new one, then the caller must
4431 * have uninitialized any derived state in the victim rule, as in step 5 in
4432 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4433 * performs steps 6 and 7 for the victim rule, most notably by calling its
4434 * ->rule_dealloc() function.
4436 * If 'error' is nonzero, then generally the operation will be rolled back:
4438 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4439 * restores the original rule. The caller must have uninitialized any
4440 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4441 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4442 * and 7 for the new rule, calling its ->rule_dealloc() function.
4444 * - If 'op' is a "modify flow" operation, ofproto restores the original
4447 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4448 * allowed to fail. It must always succeed.
4450 * Please see the large comment in ofproto/ofproto-provider.h titled
4451 * "Asynchronous Operation Support" for more information. */
4453 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4455 struct ofopgroup *group = op->group;
4457 ovs_assert(op->rule->pending == op);
4458 ovs_assert(group->n_running > 0);
4459 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4462 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4463 ofopgroup_complete(group);
4468 ofoperation_get_victim(struct ofoperation *op)
4470 ovs_assert(op->type == OFOPERATION_ADD);
4475 pick_datapath_id(const struct ofproto *ofproto)
4477 const struct ofport *port;
4479 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4481 uint8_t ea[ETH_ADDR_LEN];
4484 error = netdev_get_etheraddr(port->netdev, ea);
4486 return eth_addr_to_uint64(ea);
4488 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4489 ofproto->name, netdev_get_name(port->netdev),
4492 return ofproto->fallback_dpid;
4496 pick_fallback_dpid(void)
4498 uint8_t ea[ETH_ADDR_LEN];
4499 eth_addr_nicira_random(ea);
4500 return eth_addr_to_uint64(ea);
4503 /* Table overflow policy. */
4505 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4506 * is not configured to evict rules or if the table contains no evictable
4507 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4509 static struct rule *
4510 choose_rule_to_evict(struct oftable *table)
4512 struct eviction_group *evg;
4514 if (!table->eviction_fields) {
4518 /* In the common case, the outer and inner loops here will each be entered
4521 * - The inner loop normally "return"s in its first iteration. If the
4522 * eviction group has any evictable rules, then it always returns in
4525 * - The outer loop only iterates more than once if the largest eviction
4526 * group has no evictable rules.
4528 * - The outer loop can exit only if table's 'max_flows' is all filled up
4529 * by unevictable rules'. */
4530 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4533 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4534 if (rule->evictable) {
4543 /* Searches 'ofproto' for tables that have more flows than their configured
4544 * maximum and that have flow eviction enabled, and evicts as many flows as
4545 * necessary and currently feasible from them.
4547 * This triggers only when an OpenFlow table has N flows in it and then the
4548 * client configures a maximum number of flows less than N. */
4550 ofproto_evict(struct ofproto *ofproto)
4552 struct ofopgroup *group;
4553 struct oftable *table;
4555 group = ofopgroup_create_unattached(ofproto);
4556 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4557 while (classifier_count(&table->cls) > table->max_flows
4558 && table->eviction_fields) {
4561 rule = choose_rule_to_evict(table);
4562 if (!rule || rule->pending) {
4566 ofoperation_create(group, rule,
4567 OFOPERATION_DELETE, OFPRR_EVICTION);
4568 oftable_remove_rule(rule);
4569 ofproto->ofproto_class->rule_destruct(rule);
4572 ofopgroup_submit(group);
4575 /* Eviction groups. */
4577 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4578 * rules. The priority contains low-order random bits to ensure that eviction
4579 * groups with the same number of rules are prioritized randomly. */
4581 eviction_group_priority(size_t n_rules)
4583 uint16_t size = MIN(UINT16_MAX, n_rules);
4584 return (size << 16) | random_uint16();
4587 /* Updates 'evg', an eviction_group within 'table', following a change that
4588 * adds or removes rules in 'evg'. */
4590 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4592 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4593 eviction_group_priority(heap_count(&evg->rules)));
4596 /* Destroys 'evg', an eviction_group within 'table':
4598 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4599 * rules themselves, just removes them from the eviction group.)
4601 * - Removes 'evg' from 'table'.
4605 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4607 while (!heap_is_empty(&evg->rules)) {
4610 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4611 rule->eviction_group = NULL;
4613 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4614 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4615 heap_destroy(&evg->rules);
4619 /* Removes 'rule' from its eviction group, if any. */
4621 eviction_group_remove_rule(struct rule *rule)
4623 if (rule->eviction_group) {
4624 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4625 struct eviction_group *evg = rule->eviction_group;
4627 rule->eviction_group = NULL;
4628 heap_remove(&evg->rules, &rule->evg_node);
4629 if (heap_is_empty(&evg->rules)) {
4630 eviction_group_destroy(table, evg);
4632 eviction_group_resized(table, evg);
4637 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4638 * returns the hash value. */
4640 eviction_group_hash_rule(struct rule *rule)
4642 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4643 const struct mf_subfield *sf;
4647 hash = table->eviction_group_id_basis;
4648 miniflow_expand(&rule->cr.match.flow, &flow);
4649 for (sf = table->eviction_fields;
4650 sf < &table->eviction_fields[table->n_eviction_fields];
4653 if (mf_are_prereqs_ok(sf->field, &flow)) {
4654 union mf_value value;
4656 mf_get_value(sf->field, &flow, &value);
4658 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4660 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4661 unsigned int start = sf->ofs + sf->n_bits;
4662 bitwise_zero(&value, sf->field->n_bytes, start,
4663 sf->field->n_bytes * 8 - start);
4665 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4667 hash = hash_int(hash, 0);
4674 /* Returns an eviction group within 'table' with the given 'id', creating one
4676 static struct eviction_group *
4677 eviction_group_find(struct oftable *table, uint32_t id)
4679 struct eviction_group *evg;
4681 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4685 evg = xmalloc(sizeof *evg);
4686 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4687 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4688 eviction_group_priority(0));
4689 heap_init(&evg->rules);
4694 /* Returns an eviction priority for 'rule'. The return value should be
4695 * interpreted so that higher priorities make a rule more attractive candidates
4698 rule_eviction_priority(struct rule *rule)
4700 long long int hard_expiration;
4701 long long int idle_expiration;
4702 long long int expiration;
4703 uint32_t expiration_offset;
4705 /* Calculate time of expiration. */
4706 hard_expiration = (rule->hard_timeout
4707 ? rule->modified + rule->hard_timeout * 1000
4709 idle_expiration = (rule->idle_timeout
4710 ? rule->used + rule->idle_timeout * 1000
4712 expiration = MIN(hard_expiration, idle_expiration);
4713 if (expiration == LLONG_MAX) {
4717 /* Calculate the time of expiration as a number of (approximate) seconds
4718 * after program startup.
4720 * This should work OK for program runs that last UINT32_MAX seconds or
4721 * less. Therefore, please restart OVS at least once every 136 years. */
4722 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4724 /* Invert the expiration offset because we're using a max-heap. */
4725 return UINT32_MAX - expiration_offset;
4728 /* Adds 'rule' to an appropriate eviction group for its oftable's
4729 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4730 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4733 * The caller must ensure that 'rule' is not already in an eviction group. */
4735 eviction_group_add_rule(struct rule *rule)
4737 struct ofproto *ofproto = rule->ofproto;
4738 struct oftable *table = &ofproto->tables[rule->table_id];
4740 if (table->eviction_fields
4741 && (rule->hard_timeout || rule->idle_timeout)) {
4742 struct eviction_group *evg;
4744 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4746 rule->eviction_group = evg;
4747 heap_insert(&evg->rules, &rule->evg_node,
4748 rule_eviction_priority(rule));
4749 eviction_group_resized(table, evg);
4755 /* Initializes 'table'. */
4757 oftable_init(struct oftable *table)
4759 memset(table, 0, sizeof *table);
4760 classifier_init(&table->cls);
4761 table->max_flows = UINT_MAX;
4764 /* Destroys 'table', including its classifier and eviction groups.
4766 * The caller is responsible for freeing 'table' itself. */
4768 oftable_destroy(struct oftable *table)
4770 ovs_assert(classifier_is_empty(&table->cls));
4771 oftable_disable_eviction(table);
4772 classifier_destroy(&table->cls);
4776 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4777 * string, then 'table' will use its default name.
4779 * This only affects the name exposed for a table exposed through the OpenFlow
4780 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4782 oftable_set_name(struct oftable *table, const char *name)
4784 if (name && name[0]) {
4785 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4786 if (!table->name || strncmp(name, table->name, len)) {
4788 table->name = xmemdup0(name, len);
4796 /* oftables support a choice of two policies when adding a rule would cause the
4797 * number of flows in the table to exceed the configured maximum number: either
4798 * they can refuse to add the new flow or they can evict some existing flow.
4799 * This function configures the former policy on 'table'. */
4801 oftable_disable_eviction(struct oftable *table)
4803 if (table->eviction_fields) {
4804 struct eviction_group *evg, *next;
4806 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4807 &table->eviction_groups_by_id) {
4808 eviction_group_destroy(table, evg);
4810 hmap_destroy(&table->eviction_groups_by_id);
4811 heap_destroy(&table->eviction_groups_by_size);
4813 free(table->eviction_fields);
4814 table->eviction_fields = NULL;
4815 table->n_eviction_fields = 0;
4819 /* oftables support a choice of two policies when adding a rule would cause the
4820 * number of flows in the table to exceed the configured maximum number: either
4821 * they can refuse to add the new flow or they can evict some existing flow.
4822 * This function configures the latter policy on 'table', with fairness based
4823 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4824 * 'n_fields' as 0 disables fairness.) */
4826 oftable_enable_eviction(struct oftable *table,
4827 const struct mf_subfield *fields, size_t n_fields)
4829 struct cls_cursor cursor;
4832 if (table->eviction_fields
4833 && n_fields == table->n_eviction_fields
4835 || !memcmp(fields, table->eviction_fields,
4836 n_fields * sizeof *fields))) {
4841 oftable_disable_eviction(table);
4843 table->n_eviction_fields = n_fields;
4844 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4846 table->eviction_group_id_basis = random_uint32();
4847 hmap_init(&table->eviction_groups_by_id);
4848 heap_init(&table->eviction_groups_by_size);
4850 cls_cursor_init(&cursor, &table->cls, NULL);
4851 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4852 eviction_group_add_rule(rule);
4856 /* Removes 'rule' from the oftable that contains it. */
4858 oftable_remove_rule(struct rule *rule)
4860 struct ofproto *ofproto = rule->ofproto;
4861 struct oftable *table = &ofproto->tables[rule->table_id];
4863 classifier_remove(&table->cls, &rule->cr);
4864 eviction_group_remove_rule(rule);
4865 if (!list_is_empty(&rule->expirable)) {
4866 list_remove(&rule->expirable);
4870 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4871 * oftable that has an identical cls_rule. Returns the rule that was removed,
4872 * if any, and otherwise NULL. */
4873 static struct rule *
4874 oftable_replace_rule(struct rule *rule)
4876 struct ofproto *ofproto = rule->ofproto;
4877 struct oftable *table = &ofproto->tables[rule->table_id];
4878 struct rule *victim;
4879 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4882 list_insert(&ofproto->expirable, &rule->expirable);
4885 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4887 if (!list_is_empty(&victim->expirable)) {
4888 list_remove(&victim->expirable);
4890 eviction_group_remove_rule(victim);
4892 eviction_group_add_rule(rule);
4896 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4898 oftable_substitute_rule(struct rule *old, struct rule *new)
4901 oftable_replace_rule(new);
4903 oftable_remove_rule(old);
4907 /* unixctl commands. */
4910 ofproto_lookup(const char *name)
4912 struct ofproto *ofproto;
4914 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4916 if (!strcmp(ofproto->name, name)) {
4924 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4925 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4927 struct ofproto *ofproto;
4931 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4932 ds_put_format(&results, "%s\n", ofproto->name);
4934 unixctl_command_reply(conn, ds_cstr(&results));
4935 ds_destroy(&results);
4939 ofproto_unixctl_init(void)
4941 static bool registered;
4947 unixctl_command_register("ofproto/list", "", 0, 0,
4948 ofproto_unixctl_list, NULL);
4951 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4953 * This is deprecated. It is only for compatibility with broken device drivers
4954 * in old versions of Linux that do not properly support VLANs when VLAN
4955 * devices are not used. When broken device drivers are no longer in
4956 * widespread use, we will delete these interfaces. */
4958 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4959 * (exactly) by an OpenFlow rule in 'ofproto'. */
4961 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4963 const struct oftable *oftable;
4965 free(ofproto->vlan_bitmap);
4966 ofproto->vlan_bitmap = bitmap_allocate(4096);
4967 ofproto->vlans_changed = false;
4969 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4970 const struct cls_table *table;
4972 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4973 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
4974 const struct cls_rule *rule;
4976 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4977 uint16_t vid = miniflow_get_vid(&rule->match.flow);
4978 bitmap_set1(vlan_bitmap, vid);
4979 bitmap_set1(ofproto->vlan_bitmap, vid);
4986 /* Returns true if new VLANs have come into use by the flow table since the
4987 * last call to ofproto_get_vlan_usage().
4989 * We don't track when old VLANs stop being used. */
4991 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4993 return ofproto->vlans_changed;
4996 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4997 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4998 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4999 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5000 * then the VLAN device is un-enslaved. */
5002 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
5003 uint16_t realdev_ofp_port, int vid)
5005 struct ofport *ofport;
5008 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5010 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5012 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5013 ofproto->name, vlandev_ofp_port);
5017 if (!ofproto->ofproto_class->set_realdev) {
5018 if (!vlandev_ofp_port) {
5021 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5025 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5027 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5028 ofproto->name, vlandev_ofp_port,
5029 netdev_get_name(ofport->netdev), strerror(error));