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 hindex_init(&ofproto->cookies);
428 list_init(&ofproto->expirable);
429 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
430 ofproto->state = S_OPENFLOW;
431 list_init(&ofproto->pending);
432 ofproto->n_pending = 0;
433 hmap_init(&ofproto->deletions);
434 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
435 ofproto->first_op = ofproto->last_op = LLONG_MIN;
436 ofproto->next_op_report = LLONG_MAX;
437 ofproto->op_backoff = LLONG_MIN;
438 ofproto->vlan_bitmap = NULL;
439 ofproto->vlans_changed = false;
440 ofproto->min_mtu = INT_MAX;
442 error = ofproto->ofproto_class->construct(ofproto);
444 VLOG_ERR("failed to open datapath %s: %s",
445 datapath_name, strerror(error));
446 ofproto_destroy__(ofproto);
450 /* The "max_ports" member should have been set by ->construct(ofproto).
451 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
452 ofproto->ofp_port_ids = bitmap_allocate(ofp_to_u16(ofproto->max_ports));
453 bitmap_set1(ofproto->ofp_port_ids, 0);
455 /* Check that hidden tables, if any, are at the end. */
456 ovs_assert(ofproto->n_tables);
457 for (i = 0; i + 1 < ofproto->n_tables; i++) {
458 enum oftable_flags flags = ofproto->tables[i].flags;
459 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
461 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
464 ofproto->datapath_id = pick_datapath_id(ofproto);
471 /* Must be called (only) by an ofproto implementation in its constructor
472 * function. See the large comment on 'construct' in struct ofproto_class for
475 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
477 struct oftable *table;
479 ovs_assert(!ofproto->n_tables);
480 ovs_assert(n_tables >= 1 && n_tables <= 255);
482 ofproto->n_tables = n_tables;
483 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
484 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
489 /* To be optionally called (only) by an ofproto implementation in its
490 * constructor function. See the large comment on 'construct' in struct
491 * ofproto_class for details.
493 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
494 * will then ensure that actions passed into the ofproto implementation will
495 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
496 * function is not called, there will be no such restriction.
498 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
499 * the 'max_ports' restriction. */
501 ofproto_init_max_ports(struct ofproto *ofproto, ofp_port_t max_ports)
503 ovs_assert(ofp_to_u16(max_ports) <= ofp_to_u16(OFPP_MAX));
504 ofproto->max_ports = max_ports;
508 ofproto_get_datapath_id(const struct ofproto *ofproto)
510 return ofproto->datapath_id;
514 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
516 uint64_t old_dpid = p->datapath_id;
517 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
518 if (p->datapath_id != old_dpid) {
519 /* Force all active connections to reconnect, since there is no way to
520 * notify a controller that the datapath ID has changed. */
521 ofproto_reconnect_controllers(p);
526 ofproto_set_controllers(struct ofproto *p,
527 const struct ofproto_controller *controllers,
528 size_t n_controllers, uint32_t allowed_versions)
530 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
535 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
537 connmgr_set_fail_mode(p->connmgr, fail_mode);
540 /* Drops the connections between 'ofproto' and all of its controllers, forcing
541 * them to reconnect. */
543 ofproto_reconnect_controllers(struct ofproto *ofproto)
545 connmgr_reconnect(ofproto->connmgr);
548 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
549 * in-band control should guarantee access, in the same way that in-band
550 * control guarantees access to OpenFlow controllers. */
552 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
553 const struct sockaddr_in *extras, size_t n)
555 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
558 /* Sets the OpenFlow queue used by flows set up by in-band control on
559 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
560 * flows will use the default queue. */
562 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
564 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
567 /* Sets the number of flows at which eviction from the kernel flow table
570 ofproto_set_flow_eviction_threshold(unsigned threshold)
572 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
576 /* If forward_bpdu is true, the NORMAL action will forward frames with
577 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
578 * the NORMAL action will drop these frames. */
580 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
582 bool old_val = ofproto->forward_bpdu;
583 ofproto->forward_bpdu = forward_bpdu;
584 if (old_val != ofproto->forward_bpdu) {
585 if (ofproto->ofproto_class->forward_bpdu_changed) {
586 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
591 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
592 * 'idle_time', in seconds, and the maximum number of MAC table entries to
595 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
598 if (ofproto->ofproto_class->set_mac_table_config) {
599 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
605 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
608 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
612 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
614 return connmgr_set_snoops(ofproto->connmgr, snoops);
618 ofproto_set_netflow(struct ofproto *ofproto,
619 const struct netflow_options *nf_options)
621 if (nf_options && sset_is_empty(&nf_options->collectors)) {
625 if (ofproto->ofproto_class->set_netflow) {
626 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
628 return nf_options ? EOPNOTSUPP : 0;
633 ofproto_set_sflow(struct ofproto *ofproto,
634 const struct ofproto_sflow_options *oso)
636 if (oso && sset_is_empty(&oso->targets)) {
640 if (ofproto->ofproto_class->set_sflow) {
641 return ofproto->ofproto_class->set_sflow(ofproto, oso);
643 return oso ? EOPNOTSUPP : 0;
648 ofproto_set_ipfix(struct ofproto *ofproto,
649 const struct ofproto_ipfix_bridge_exporter_options *bo,
650 const struct ofproto_ipfix_flow_exporter_options *fo,
653 if (ofproto->ofproto_class->set_ipfix) {
654 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
656 return (bo || fo) ? EOPNOTSUPP : 0;
661 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
663 flow_restore_wait = flow_restore_wait_db;
667 ofproto_get_flow_restore_wait(void)
669 return flow_restore_wait;
673 /* Spanning Tree Protocol (STP) configuration. */
675 /* Configures STP on 'ofproto' using the settings defined in 's'. If
676 * 's' is NULL, disables STP.
678 * Returns 0 if successful, otherwise a positive errno value. */
680 ofproto_set_stp(struct ofproto *ofproto,
681 const struct ofproto_stp_settings *s)
683 return (ofproto->ofproto_class->set_stp
684 ? ofproto->ofproto_class->set_stp(ofproto, s)
688 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
689 * 'enabled' member of 's' is false, then the other members are not
692 * Returns 0 if successful, otherwise a positive errno value. */
694 ofproto_get_stp_status(struct ofproto *ofproto,
695 struct ofproto_stp_status *s)
697 return (ofproto->ofproto_class->get_stp_status
698 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
702 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
703 * in 's'. The caller is responsible for assigning STP port numbers
704 * (using the 'port_num' member in the range of 1 through 255, inclusive)
705 * and ensuring there are no duplicates. If the 's' is NULL, then STP
706 * is disabled on the port.
708 * Returns 0 if successful, otherwise a positive errno value.*/
710 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
711 const struct ofproto_port_stp_settings *s)
713 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
715 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
716 ofproto->name, ofp_port);
720 return (ofproto->ofproto_class->set_stp_port
721 ? ofproto->ofproto_class->set_stp_port(ofport, s)
725 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
726 * 's'. If the 'enabled' member in 's' is false, then the other members
727 * are not meaningful.
729 * Returns 0 if successful, otherwise a positive errno value.*/
731 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
732 struct ofproto_port_stp_status *s)
734 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
736 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
737 "port %"PRIu16, ofproto->name, ofp_port);
741 return (ofproto->ofproto_class->get_stp_port_status
742 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
746 /* Queue DSCP configuration. */
748 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
749 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
750 * to implement QoS. Instead, it is used to implement features which require
751 * knowledge of what queues exist on a port, and some basic information about
754 * Returns 0 if successful, otherwise a positive errno value. */
756 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
757 const struct ofproto_port_queue *queues,
760 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
763 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
764 ofproto->name, ofp_port);
768 return (ofproto->ofproto_class->set_queues
769 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
773 /* Connectivity Fault Management configuration. */
775 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
777 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
779 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
780 if (ofport && ofproto->ofproto_class->set_cfm) {
781 ofproto->ofproto_class->set_cfm(ofport, NULL);
785 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
786 * basic configuration from the configuration members in 'cfm', and the remote
787 * maintenance point ID from remote_mpid. Ignores the statistics members of
790 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
792 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
793 const struct cfm_settings *s)
795 struct ofport *ofport;
798 ofport = ofproto_get_port(ofproto, ofp_port);
800 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
801 ofproto->name, ofp_port);
805 /* XXX: For configuration simplicity, we only support one remote_mpid
806 * outside of the CFM module. It's not clear if this is the correct long
807 * term solution or not. */
808 error = (ofproto->ofproto_class->set_cfm
809 ? ofproto->ofproto_class->set_cfm(ofport, s)
812 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
813 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
818 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
819 * 'ofproto' does not have a port 'ofp_port'. */
821 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
822 const struct smap *cfg)
824 struct ofport *ofport;
827 ofport = ofproto_get_port(ofproto, ofp_port);
829 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
830 ofproto->name, ofp_port);
834 error = (ofproto->ofproto_class->set_bfd
835 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
838 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
839 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
844 /* Populates 'status' with key value pairs indicating the status of the BFD
845 * session on 'ofp_port'. This information is intended to be populated in the
846 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
849 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
852 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
853 return (ofport && ofproto->ofproto_class->get_bfd_status
854 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
858 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
859 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
860 * 0 if LACP partner information is not current (generally indicating a
861 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
863 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
865 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
866 return (ofport && ofproto->ofproto_class->port_is_lacp_current
867 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
873 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
874 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
875 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
876 * configuration plus, if there is more than one slave, a bonding
879 * If 'aux' is already registered then this function updates its configuration
880 * to 's'. Otherwise, this function registers a new bundle.
882 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
885 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
886 const struct ofproto_bundle_settings *s)
888 return (ofproto->ofproto_class->bundle_set
889 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
893 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
894 * If no such bundle has been registered, this has no effect. */
896 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
898 return ofproto_bundle_register(ofproto, aux, NULL);
902 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
903 * If 'aux' is already registered then this function updates its configuration
904 * to 's'. Otherwise, this function registers a new mirror. */
906 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
907 const struct ofproto_mirror_settings *s)
909 return (ofproto->ofproto_class->mirror_set
910 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
914 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
915 * If no mirror has been registered, this has no effect. */
917 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
919 return ofproto_mirror_register(ofproto, aux, NULL);
922 /* Retrieves statistics from mirror associated with client data pointer
923 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
924 * 'bytes', respectively. If a particular counters is not supported,
925 * the appropriate argument is set to UINT64_MAX. */
927 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
928 uint64_t *packets, uint64_t *bytes)
930 if (!ofproto->ofproto_class->mirror_get_stats) {
931 *packets = *bytes = UINT64_MAX;
935 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
939 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
940 * which all packets are flooded, instead of using MAC learning. If
941 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
943 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
946 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
948 return (ofproto->ofproto_class->set_flood_vlans
949 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
953 /* Returns true if 'aux' is a registered bundle that is currently in use as the
954 * output for a mirror. */
956 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
958 return (ofproto->ofproto_class->is_mirror_output_bundle
959 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
963 /* Configuration of OpenFlow tables. */
965 /* Returns the number of OpenFlow tables in 'ofproto'. */
967 ofproto_get_n_tables(const struct ofproto *ofproto)
969 return ofproto->n_tables;
972 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
973 * settings from 's'. 'table_id' must be in the range 0 through the number of
974 * OpenFlow tables in 'ofproto' minus 1, inclusive.
976 * For read-only tables, only the name may be configured. */
978 ofproto_configure_table(struct ofproto *ofproto, int table_id,
979 const struct ofproto_table_settings *s)
981 struct oftable *table;
983 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
984 table = &ofproto->tables[table_id];
986 oftable_set_name(table, s->name);
988 if (table->flags & OFTABLE_READONLY) {
993 oftable_enable_eviction(table, s->groups, s->n_groups);
995 oftable_disable_eviction(table);
998 table->max_flows = s->max_flows;
999 if (classifier_count(&table->cls) > table->max_flows
1000 && table->eviction_fields) {
1001 /* 'table' contains more flows than allowed. We might not be able to
1002 * evict them right away because of the asynchronous nature of flow
1003 * table changes. Schedule eviction for later. */
1004 switch (ofproto->state) {
1006 ofproto->state = S_EVICT;
1010 /* We're already deleting flows, nothing more to do. */
1017 ofproto_has_snoops(const struct ofproto *ofproto)
1019 return connmgr_has_snoops(ofproto->connmgr);
1023 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1025 connmgr_get_snoops(ofproto->connmgr, snoops);
1029 ofproto_flush__(struct ofproto *ofproto)
1031 struct ofopgroup *group;
1032 struct oftable *table;
1034 if (ofproto->ofproto_class->flush) {
1035 ofproto->ofproto_class->flush(ofproto);
1038 group = ofopgroup_create_unattached(ofproto);
1039 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1040 struct rule *rule, *next_rule;
1041 struct cls_cursor cursor;
1043 if (table->flags & OFTABLE_HIDDEN) {
1047 cls_cursor_init(&cursor, &table->cls, NULL);
1048 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1049 if (!rule->pending) {
1050 ofoperation_create(group, rule, OFOPERATION_DELETE,
1052 oftable_remove_rule(rule);
1053 ofproto->ofproto_class->rule_destruct(rule);
1057 ofopgroup_submit(group);
1061 ofproto_destroy__(struct ofproto *ofproto)
1063 struct oftable *table;
1065 ovs_assert(list_is_empty(&ofproto->pending));
1066 ovs_assert(!ofproto->n_pending);
1068 connmgr_destroy(ofproto->connmgr);
1070 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1071 free(ofproto->name);
1072 free(ofproto->type);
1073 free(ofproto->mfr_desc);
1074 free(ofproto->hw_desc);
1075 free(ofproto->sw_desc);
1076 free(ofproto->serial_desc);
1077 free(ofproto->dp_desc);
1078 hmap_destroy(&ofproto->ports);
1079 shash_destroy(&ofproto->port_by_name);
1080 bitmap_free(ofproto->ofp_port_ids);
1081 simap_destroy(&ofproto->ofp_requests);
1083 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1084 oftable_destroy(table);
1086 free(ofproto->tables);
1088 hmap_destroy(&ofproto->deletions);
1090 free(ofproto->vlan_bitmap);
1092 ofproto->ofproto_class->dealloc(ofproto);
1096 ofproto_destroy(struct ofproto *p)
1098 struct ofport *ofport, *next_ofport;
1105 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1106 ofport_destroy(ofport);
1109 p->ofproto_class->destruct(p);
1110 ofproto_destroy__(p);
1113 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1114 * kernel datapath, for example, this destroys the datapath in the kernel, and
1115 * with the netdev-based datapath, it tears down the data structures that
1116 * represent the datapath.
1118 * The datapath should not be currently open as an ofproto. */
1120 ofproto_delete(const char *name, const char *type)
1122 const struct ofproto_class *class = ofproto_class_find__(type);
1123 return (!class ? EAFNOSUPPORT
1124 : !class->del ? EACCES
1125 : class->del(type, name));
1129 process_port_change(struct ofproto *ofproto, int error, char *devname)
1131 if (error == ENOBUFS) {
1132 reinit_ports(ofproto);
1133 } else if (!error) {
1134 update_port(ofproto, devname);
1140 ofproto_type_run(const char *datapath_type)
1142 const struct ofproto_class *class;
1145 datapath_type = ofproto_normalize_type(datapath_type);
1146 class = ofproto_class_find__(datapath_type);
1148 error = class->type_run ? class->type_run(datapath_type) : 0;
1149 if (error && error != EAGAIN) {
1150 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1151 datapath_type, strerror(error));
1157 ofproto_type_run_fast(const char *datapath_type)
1159 const struct ofproto_class *class;
1162 datapath_type = ofproto_normalize_type(datapath_type);
1163 class = ofproto_class_find__(datapath_type);
1165 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1166 if (error && error != EAGAIN) {
1167 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1168 datapath_type, strerror(error));
1174 ofproto_type_wait(const char *datapath_type)
1176 const struct ofproto_class *class;
1178 datapath_type = ofproto_normalize_type(datapath_type);
1179 class = ofproto_class_find__(datapath_type);
1181 if (class->type_wait) {
1182 class->type_wait(datapath_type);
1187 ofproto_run(struct ofproto *p)
1189 struct sset changed_netdevs;
1190 const char *changed_netdev;
1191 struct ofport *ofport;
1194 error = p->ofproto_class->run(p);
1195 if (error && error != EAGAIN) {
1196 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1199 if (p->ofproto_class->port_poll) {
1202 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1203 process_port_change(p, error, devname);
1207 /* Update OpenFlow port status for any port whose netdev has changed.
1209 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1210 * destroyed, so it's not safe to update ports directly from the
1211 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1212 * need this two-phase approach. */
1213 sset_init(&changed_netdevs);
1214 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1215 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1216 if (ofport->change_seq != change_seq) {
1217 ofport->change_seq = change_seq;
1218 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1221 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1222 update_port(p, changed_netdev);
1224 sset_destroy(&changed_netdevs);
1228 connmgr_run(p->connmgr, handle_openflow);
1232 connmgr_run(p->connmgr, NULL);
1234 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1235 p->state = S_OPENFLOW;
1240 connmgr_run(p->connmgr, NULL);
1242 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1243 connmgr_flushed(p->connmgr);
1244 p->state = S_OPENFLOW;
1252 if (time_msec() >= p->next_op_report) {
1253 long long int ago = (time_msec() - p->first_op) / 1000;
1254 long long int interval = (p->last_op - p->first_op) / 1000;
1258 ds_put_format(&s, "%d flow_mods ",
1259 p->n_add + p->n_delete + p->n_modify);
1260 if (interval == ago) {
1261 ds_put_format(&s, "in the last %lld s", ago);
1262 } else if (interval) {
1263 ds_put_format(&s, "in the %lld s starting %lld s ago",
1266 ds_put_format(&s, "%lld s ago", ago);
1269 ds_put_cstr(&s, " (");
1271 ds_put_format(&s, "%d adds, ", p->n_add);
1274 ds_put_format(&s, "%d deletes, ", p->n_delete);
1277 ds_put_format(&s, "%d modifications, ", p->n_modify);
1280 ds_put_char(&s, ')');
1282 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1285 p->n_add = p->n_delete = p->n_modify = 0;
1286 p->next_op_report = LLONG_MAX;
1292 /* Performs periodic activity required by 'ofproto' that needs to be done
1293 * with the least possible latency.
1295 * It makes sense to call this function a couple of times per poll loop, to
1296 * provide a significant performance boost on some benchmarks with the
1297 * ofproto-dpif implementation. */
1299 ofproto_run_fast(struct ofproto *p)
1303 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1304 if (error && error != EAGAIN) {
1305 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1306 p->name, strerror(error));
1312 ofproto_wait(struct ofproto *p)
1314 struct ofport *ofport;
1316 p->ofproto_class->wait(p);
1317 if (p->ofproto_class->port_poll_wait) {
1318 p->ofproto_class->port_poll_wait(p);
1321 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1322 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1323 poll_immediate_wake();
1329 connmgr_wait(p->connmgr, true);
1334 connmgr_wait(p->connmgr, false);
1335 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1336 poll_immediate_wake();
1343 ofproto_is_alive(const struct ofproto *p)
1345 return connmgr_has_controllers(p->connmgr);
1348 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1349 * memory_report(). */
1351 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1353 const struct oftable *table;
1354 unsigned int n_rules;
1356 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1357 simap_increase(usage, "ops",
1358 ofproto->n_pending + hmap_count(&ofproto->deletions));
1361 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1362 n_rules += classifier_count(&table->cls);
1364 simap_increase(usage, "rules", n_rules);
1366 if (ofproto->ofproto_class->get_memory_usage) {
1367 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1370 connmgr_get_memory_usage(ofproto->connmgr, usage);
1374 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1377 connmgr_get_controller_info(ofproto->connmgr, info);
1381 ofproto_free_ofproto_controller_info(struct shash *info)
1383 connmgr_free_controller_info(info);
1386 /* Makes a deep copy of 'old' into 'port'. */
1388 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1390 port->name = xstrdup(old->name);
1391 port->type = xstrdup(old->type);
1392 port->ofp_port = old->ofp_port;
1395 /* Frees memory allocated to members of 'ofproto_port'.
1397 * Do not call this function on an ofproto_port obtained from
1398 * ofproto_port_dump_next(): that function retains ownership of the data in the
1401 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1403 free(ofproto_port->name);
1404 free(ofproto_port->type);
1407 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1409 * This function provides no status indication. An error status for the entire
1410 * dump operation is provided when it is completed by calling
1411 * ofproto_port_dump_done().
1414 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1415 const struct ofproto *ofproto)
1417 dump->ofproto = ofproto;
1418 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1422 /* Attempts to retrieve another port from 'dump', which must have been created
1423 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1424 * 'port' and returns true. On failure, returns false.
1426 * Failure might indicate an actual error or merely that the last port has been
1427 * dumped. An error status for the entire dump operation is provided when it
1428 * is completed by calling ofproto_port_dump_done().
1430 * The ofproto owns the data stored in 'port'. It will remain valid until at
1431 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1432 * ofproto_port_dump_done(). */
1434 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1435 struct ofproto_port *port)
1437 const struct ofproto *ofproto = dump->ofproto;
1443 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1446 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1452 /* Completes port table dump operation 'dump', which must have been created
1453 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1454 * error-free, otherwise a positive errno value describing the problem. */
1456 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1458 const struct ofproto *ofproto = dump->ofproto;
1460 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1463 return dump->error == EOF ? 0 : dump->error;
1466 /* Returns the type to pass to netdev_open() when a datapath of type
1467 * 'datapath_type' has a port of type 'port_type', for a few special
1468 * cases when a netdev type differs from a port type. For example, when
1469 * using the userspace datapath, a port of type "internal" needs to be
1472 * Returns either 'type' itself or a string literal, which must not be
1475 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1477 const struct ofproto_class *class;
1479 datapath_type = ofproto_normalize_type(datapath_type);
1480 class = ofproto_class_find__(datapath_type);
1485 return (class->port_open_type
1486 ? class->port_open_type(datapath_type, port_type)
1490 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1491 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1492 * the port's OpenFlow port number.
1494 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1495 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1496 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1497 * 'ofp_portp' is non-null). */
1499 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1500 ofp_port_t *ofp_portp)
1502 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1505 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1507 const char *netdev_name = netdev_get_name(netdev);
1509 simap_put(&ofproto->ofp_requests, netdev_name,
1510 ofp_to_u16(ofp_port));
1511 update_port(ofproto, netdev_name);
1514 struct ofproto_port ofproto_port;
1516 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1518 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1519 ofproto_port_destroy(&ofproto_port);
1524 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1525 * initializes '*port' appropriately; on failure, returns a positive errno
1528 * The caller owns the data in 'ofproto_port' and must free it with
1529 * ofproto_port_destroy() when it is no longer needed. */
1531 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1532 struct ofproto_port *port)
1536 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1538 memset(port, 0, sizeof *port);
1543 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1544 * Returns 0 if successful, otherwise a positive errno. */
1546 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1548 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1549 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1550 struct simap_node *ofp_request_node;
1553 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1554 if (ofp_request_node) {
1555 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1558 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1559 if (!error && ofport) {
1560 /* 'name' is the netdev's name and update_port() is going to close the
1561 * netdev. Just in case update_port() refers to 'name' after it
1562 * destroys 'ofport', make a copy of it around the update_port()
1564 char *devname = xstrdup(name);
1565 update_port(ofproto, devname);
1571 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1572 * performs the 'n_actions' actions in 'actions'. The new flow will not
1575 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1576 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1577 * controllers; otherwise, it will be hidden.
1579 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1581 * This is a helper function for in-band control and fail-open. */
1583 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1584 unsigned int priority,
1585 const struct ofpact *ofpacts, size_t ofpacts_len)
1587 const struct rule *rule;
1589 rule = rule_from_cls_rule(classifier_find_match_exactly(
1590 &ofproto->tables[0].cls, match, priority));
1591 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1592 ofpacts, ofpacts_len)) {
1593 struct ofputil_flow_mod fm;
1595 memset(&fm, 0, sizeof fm);
1597 fm.priority = priority;
1598 fm.buffer_id = UINT32_MAX;
1599 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1600 fm.ofpacts_len = ofpacts_len;
1601 add_flow(ofproto, NULL, &fm, NULL);
1606 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1607 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1608 * operation cannot be initiated now but may be retried later.
1610 * This is a helper function for in-band control and fail-open. */
1612 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1614 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1617 /* Searches for a rule with matching criteria exactly equal to 'target' in
1618 * ofproto's table 0 and, if it finds one, deletes it.
1620 * This is a helper function for in-band control and fail-open. */
1622 ofproto_delete_flow(struct ofproto *ofproto,
1623 const struct match *target, unsigned int priority)
1627 rule = rule_from_cls_rule(classifier_find_match_exactly(
1628 &ofproto->tables[0].cls, target, priority));
1630 /* No such rule -> success. */
1632 } else if (rule->pending) {
1633 /* An operation on the rule is already pending -> failure.
1634 * Caller must retry later if it's important. */
1637 /* Initiate deletion -> success. */
1638 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1639 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1640 oftable_remove_rule(rule);
1641 ofproto->ofproto_class->rule_destruct(rule);
1642 ofopgroup_submit(group);
1648 /* Starts the process of deleting all of the flows from all of ofproto's flow
1649 * tables and then reintroducing the flows required by in-band control and
1650 * fail-open. The process will complete in a later call to ofproto_run(). */
1652 ofproto_flush_flows(struct ofproto *ofproto)
1654 COVERAGE_INC(ofproto_flush);
1655 ofproto->state = S_FLUSH;
1659 reinit_ports(struct ofproto *p)
1661 struct ofproto_port_dump dump;
1662 struct sset devnames;
1663 struct ofport *ofport;
1664 struct ofproto_port ofproto_port;
1665 const char *devname;
1667 COVERAGE_INC(ofproto_reinit_ports);
1669 sset_init(&devnames);
1670 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1671 sset_add(&devnames, netdev_get_name(ofport->netdev));
1673 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1674 sset_add(&devnames, ofproto_port.name);
1677 SSET_FOR_EACH (devname, &devnames) {
1678 update_port(p, devname);
1680 sset_destroy(&devnames);
1684 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1686 uint16_t max_ports = ofp_to_u16(ofproto->max_ports);
1689 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1691 port_idx = UINT16_MAX;
1694 if (port_idx >= max_ports
1695 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1696 uint16_t end_port_no = ofp_to_u16(ofproto->alloc_port_no);
1697 uint16_t alloc_port_no = end_port_no;
1699 /* Search for a free OpenFlow port number. We try not to
1700 * immediately reuse them to prevent problems due to old
1703 if (++alloc_port_no >= max_ports) {
1706 if (!bitmap_is_set(ofproto->ofp_port_ids, alloc_port_no)) {
1707 port_idx = alloc_port_no;
1708 ofproto->alloc_port_no = u16_to_ofp(alloc_port_no);
1711 if (alloc_port_no == end_port_no) {
1716 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1717 return u16_to_ofp(port_idx);
1721 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1723 if (ofp_to_u16(ofp_port) < ofp_to_u16(ofproto->max_ports)) {
1724 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1728 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1729 * pointer if the netdev cannot be opened. On success, also fills in
1731 static struct netdev *
1732 ofport_open(struct ofproto *ofproto,
1733 struct ofproto_port *ofproto_port,
1734 struct ofputil_phy_port *pp)
1736 enum netdev_flags flags;
1737 struct netdev *netdev;
1740 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1742 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1743 "cannot be opened (%s)",
1745 ofproto_port->name, ofproto_port->ofp_port,
1746 ofproto_port->name, strerror(error));
1750 if (ofproto_port->ofp_port == OFPP_NONE) {
1751 if (!strcmp(ofproto->name, ofproto_port->name)) {
1752 ofproto_port->ofp_port = OFPP_LOCAL;
1754 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
1755 ofproto_port->name);
1758 pp->port_no = ofproto_port->ofp_port;
1759 netdev_get_etheraddr(netdev, pp->hw_addr);
1760 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1761 netdev_get_flags(netdev, &flags);
1762 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1763 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1764 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1765 &pp->supported, &pp->peer);
1766 pp->curr_speed = netdev_features_to_bps(pp->curr, 0);
1767 pp->max_speed = netdev_features_to_bps(pp->supported, 0);
1772 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1773 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1776 ofport_equal(const struct ofputil_phy_port *a,
1777 const struct ofputil_phy_port *b)
1779 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1780 && a->state == b->state
1781 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1782 && a->curr == b->curr
1783 && a->advertised == b->advertised
1784 && a->supported == b->supported
1785 && a->peer == b->peer
1786 && a->curr_speed == b->curr_speed
1787 && a->max_speed == b->max_speed);
1790 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1791 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1792 * one with the same name or port number). */
1794 ofport_install(struct ofproto *p,
1795 struct netdev *netdev, const struct ofputil_phy_port *pp)
1797 const char *netdev_name = netdev_get_name(netdev);
1798 struct ofport *ofport;
1801 /* Create ofport. */
1802 ofport = p->ofproto_class->port_alloc();
1807 ofport->ofproto = p;
1808 ofport->netdev = netdev;
1809 ofport->change_seq = netdev_change_seq(netdev);
1811 ofport->ofp_port = pp->port_no;
1812 ofport->created = time_msec();
1814 /* Add port to 'p'. */
1815 hmap_insert(&p->ports, &ofport->hmap_node,
1816 hash_int(ofp_to_u16(ofport->ofp_port), 0));
1817 shash_add(&p->port_by_name, netdev_name, ofport);
1819 update_mtu(p, ofport);
1821 /* Let the ofproto_class initialize its private data. */
1822 error = p->ofproto_class->port_construct(ofport);
1826 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1830 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1831 p->name, netdev_name, strerror(error));
1833 ofport_destroy__(ofport);
1835 netdev_close(netdev);
1839 /* Removes 'ofport' from 'p' and destroys it. */
1841 ofport_remove(struct ofport *ofport)
1843 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1845 ofport_destroy(ofport);
1848 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1851 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1853 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1855 ofport_remove(port);
1859 /* Updates 'port' with new 'pp' description.
1861 * Does not handle a name or port number change. The caller must implement
1862 * such a change as a delete followed by an add. */
1864 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1866 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1867 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1868 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1869 port->pp.state = pp->state;
1870 port->pp.curr = pp->curr;
1871 port->pp.advertised = pp->advertised;
1872 port->pp.supported = pp->supported;
1873 port->pp.peer = pp->peer;
1874 port->pp.curr_speed = pp->curr_speed;
1875 port->pp.max_speed = pp->max_speed;
1877 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1880 /* Update OpenFlow 'state' in 'port' and notify controller. */
1882 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1884 if (port->pp.state != state) {
1885 port->pp.state = state;
1886 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1892 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
1894 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1896 if (port->ofproto->ofproto_class->set_realdev) {
1897 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1899 if (port->ofproto->ofproto_class->set_stp_port) {
1900 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1902 if (port->ofproto->ofproto_class->set_cfm) {
1903 port->ofproto->ofproto_class->set_cfm(port, NULL);
1905 if (port->ofproto->ofproto_class->bundle_remove) {
1906 port->ofproto->ofproto_class->bundle_remove(port);
1912 ofport_destroy__(struct ofport *port)
1914 struct ofproto *ofproto = port->ofproto;
1915 const char *name = netdev_get_name(port->netdev);
1917 hmap_remove(&ofproto->ports, &port->hmap_node);
1918 shash_delete(&ofproto->port_by_name,
1919 shash_find(&ofproto->port_by_name, name));
1921 netdev_close(port->netdev);
1922 ofproto->ofproto_class->port_dealloc(port);
1926 ofport_destroy(struct ofport *port)
1929 dealloc_ofp_port(port->ofproto, port->ofp_port);
1930 port->ofproto->ofproto_class->port_destruct(port);
1931 ofport_destroy__(port);
1936 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1938 struct ofport *port;
1940 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1941 hash_int(ofp_to_u16(ofp_port), 0),
1943 if (port->ofp_port == ofp_port) {
1951 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1953 struct ofproto *ofproto = port->ofproto;
1956 if (ofproto->ofproto_class->port_get_stats) {
1957 error = ofproto->ofproto_class->port_get_stats(port, stats);
1966 update_port(struct ofproto *ofproto, const char *name)
1968 struct ofproto_port ofproto_port;
1969 struct ofputil_phy_port pp;
1970 struct netdev *netdev;
1971 struct ofport *port;
1973 COVERAGE_INC(ofproto_update_port);
1975 /* Fetch 'name''s location and properties from the datapath. */
1976 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1977 ? ofport_open(ofproto, &ofproto_port, &pp)
1981 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1982 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1983 struct netdev *old_netdev = port->netdev;
1985 /* 'name' hasn't changed location. Any properties changed? */
1986 if (!ofport_equal(&port->pp, &pp)) {
1987 ofport_modified(port, &pp);
1990 update_mtu(ofproto, port);
1992 /* Install the newly opened netdev in case it has changed.
1993 * Don't close the old netdev yet in case port_modified has to
1994 * remove a retained reference to it.*/
1995 port->netdev = netdev;
1996 port->change_seq = netdev_change_seq(netdev);
1998 if (port->ofproto->ofproto_class->port_modified) {
1999 port->ofproto->ofproto_class->port_modified(port);
2002 netdev_close(old_netdev);
2004 /* If 'port' is nonnull then its name differs from 'name' and thus
2005 * we should delete it. If we think there's a port named 'name'
2006 * then its port number must be wrong now so delete it too. */
2008 ofport_remove(port);
2010 ofport_remove_with_name(ofproto, name);
2011 ofport_install(ofproto, netdev, &pp);
2014 /* Any port named 'name' is gone now. */
2015 ofport_remove_with_name(ofproto, name);
2017 ofproto_port_destroy(&ofproto_port);
2021 init_ports(struct ofproto *p)
2023 struct ofproto_port_dump dump;
2024 struct ofproto_port ofproto_port;
2025 struct shash_node *node, *next;
2027 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2028 const char *name = ofproto_port.name;
2030 if (shash_find(&p->port_by_name, name)) {
2031 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2034 struct ofputil_phy_port pp;
2035 struct netdev *netdev;
2037 /* Check if an OpenFlow port number had been requested. */
2038 node = shash_find(&init_ofp_ports, name);
2040 const struct iface_hint *iface_hint = node->data;
2041 simap_put(&p->ofp_requests, name,
2042 ofp_to_u16(iface_hint->ofp_port));
2045 netdev = ofport_open(p, &ofproto_port, &pp);
2047 ofport_install(p, netdev, &pp);
2052 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2053 struct iface_hint *iface_hint = node->data;
2055 if (!strcmp(iface_hint->br_name, p->name)) {
2056 free(iface_hint->br_name);
2057 free(iface_hint->br_type);
2059 shash_delete(&init_ofp_ports, node);
2066 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2067 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2069 find_min_mtu(struct ofproto *p)
2071 struct ofport *ofport;
2074 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2075 struct netdev *netdev = ofport->netdev;
2078 /* Skip any internal ports, since that's what we're trying to
2080 if (!strcmp(netdev_get_type(netdev), "internal")) {
2084 if (netdev_get_mtu(netdev, &dev_mtu)) {
2087 if (!mtu || dev_mtu < mtu) {
2092 return mtu ? mtu: ETH_PAYLOAD_MAX;
2095 /* Update MTU of all datapath devices on 'p' to the minimum of the
2096 * non-datapath ports in event of 'port' added or changed. */
2098 update_mtu(struct ofproto *p, struct ofport *port)
2100 struct ofport *ofport;
2101 struct netdev *netdev = port->netdev;
2102 int dev_mtu, old_min;
2104 if (netdev_get_mtu(netdev, &dev_mtu)) {
2108 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2109 if (dev_mtu > p->min_mtu) {
2110 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2111 dev_mtu = p->min_mtu;
2114 port->mtu = dev_mtu;
2118 /* For non-internal port find new min mtu. */
2119 old_min = p->min_mtu;
2120 port->mtu = dev_mtu;
2121 p->min_mtu = find_min_mtu(p);
2122 if (p->min_mtu == old_min) {
2126 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2127 struct netdev *netdev = ofport->netdev;
2129 if (!strcmp(netdev_get_type(netdev), "internal")) {
2130 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2131 ofport->mtu = p->min_mtu;
2138 ofproto_rule_destroy__(struct rule *rule)
2141 cls_rule_destroy(&rule->cr);
2142 free(rule->ofpacts);
2143 rule->ofproto->ofproto_class->rule_dealloc(rule);
2147 /* This function allows an ofproto implementation to destroy any rules that
2148 * remain when its ->destruct() function is called. The caller must have
2149 * already uninitialized any derived members of 'rule' (step 5 described in the
2150 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
2151 * This function implements steps 6 and 7.
2153 * This function should only be called from an ofproto implementation's
2154 * ->destruct() function. It is not suitable elsewhere. */
2156 ofproto_rule_destroy(struct rule *rule)
2158 ovs_assert(!rule->pending);
2159 oftable_remove_rule(rule);
2160 ofproto_rule_destroy__(rule);
2163 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2164 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2166 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2168 return (port == OFPP_ANY
2169 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
2172 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2173 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2175 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2177 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2182 case OFOPERATION_ADD:
2183 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
2185 case OFOPERATION_DELETE:
2188 case OFOPERATION_MODIFY:
2189 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
2195 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2196 * statistics appropriately. 'packet' must have at least sizeof(struct
2197 * ofp10_packet_in) bytes of headroom.
2199 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2200 * with statistics for 'packet' either way.
2202 * Takes ownership of 'packet'. */
2204 rule_execute(struct rule *rule, ofp_port_t in_port, struct ofpbuf *packet)
2207 union flow_in_port in_port_;
2209 ovs_assert(ofpbuf_headroom(packet) >= sizeof(struct ofp10_packet_in));
2211 in_port_.ofp_port = in_port;
2212 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
2213 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
2216 /* Returns true if 'rule' should be hidden from the controller.
2218 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2219 * (e.g. by in-band control) and are intentionally hidden from the
2222 ofproto_rule_is_hidden(const struct rule *rule)
2224 return rule->cr.priority > UINT16_MAX;
2227 static enum oftable_flags
2228 rule_get_flags(const struct rule *rule)
2230 return rule->ofproto->tables[rule->table_id].flags;
2234 rule_is_modifiable(const struct rule *rule)
2236 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2240 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2242 ofconn_send_reply(ofconn, make_echo_reply(oh));
2247 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2249 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2250 struct ofputil_switch_features features;
2251 struct ofport *port;
2257 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2259 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2261 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2262 * if present, are always at the end.) */
2263 n_tables = ofproto->n_tables;
2264 for (i = 0; i < ofproto->n_tables; i++) {
2265 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2271 features.datapath_id = ofproto->datapath_id;
2272 features.n_buffers = pktbuf_capacity();
2273 features.n_tables = n_tables;
2274 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2275 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2277 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2279 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2280 features.auxiliary_id = 0;
2281 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2283 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2284 ofputil_put_switch_features_port(&port->pp, b);
2287 ofconn_send_reply(ofconn, b);
2292 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2294 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2295 struct ofp_switch_config *osc;
2296 enum ofp_config_flags flags;
2300 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2301 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2302 flags = ofproto->frag_handling;
2303 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2304 if (oh->version < OFP13_VERSION
2305 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2306 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2308 osc->flags = htons(flags);
2309 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2310 ofconn_send_reply(ofconn, buf);
2316 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2318 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2319 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2320 uint16_t flags = ntohs(osc->flags);
2322 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2323 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2324 enum ofp_config_flags cur = ofproto->frag_handling;
2325 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2327 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2329 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2330 ofproto->frag_handling = next;
2332 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2334 ofputil_frag_handling_to_string(next));
2338 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2339 ofconn_set_invalid_ttl_to_controller(ofconn,
2340 (oh->version < OFP13_VERSION
2341 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2343 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2348 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2349 * error message code for the caller to propagate upward. Otherwise, returns
2352 * The log message mentions 'msg_type'. */
2354 reject_slave_controller(struct ofconn *ofconn)
2356 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2357 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2358 return OFPERR_OFPBRC_EPERM;
2365 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2367 struct ofproto *p = ofconn_get_ofproto(ofconn);
2368 struct ofputil_packet_out po;
2369 struct ofpbuf *payload;
2370 uint64_t ofpacts_stub[1024 / 8];
2371 struct ofpbuf ofpacts;
2373 union flow_in_port in_port_;
2376 COVERAGE_INC(ofproto_packet_out);
2378 error = reject_slave_controller(ofconn);
2383 /* Decode message. */
2384 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2385 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2387 goto exit_free_ofpacts;
2389 if (ofp_to_u16(po.in_port) >= ofp_to_u16(p->max_ports)
2390 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2391 error = OFPERR_OFPBRC_BAD_PORT;
2392 goto exit_free_ofpacts;
2397 if (po.buffer_id != UINT32_MAX) {
2398 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2399 if (error || !payload) {
2400 goto exit_free_ofpacts;
2403 payload = xmalloc(sizeof *payload);
2404 ofpbuf_use_const(payload, po.packet, po.packet_len);
2407 /* Verify actions against packet, then send packet if successful. */
2408 in_port_.ofp_port = po.in_port;
2409 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2410 error = ofpacts_check(po.ofpacts, po.ofpacts_len, &flow, p->max_ports);
2412 error = p->ofproto_class->packet_out(p, payload, &flow,
2413 po.ofpacts, po.ofpacts_len);
2415 ofpbuf_delete(payload);
2418 ofpbuf_uninit(&ofpacts);
2424 update_port_config(struct ofport *port,
2425 enum ofputil_port_config config,
2426 enum ofputil_port_config mask)
2428 enum ofputil_port_config old_config = port->pp.config;
2429 enum ofputil_port_config toggle;
2431 toggle = (config ^ port->pp.config) & mask;
2432 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2433 if (config & OFPUTIL_PC_PORT_DOWN) {
2434 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2436 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2438 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2441 port->pp.config ^= toggle;
2442 if (port->pp.config != old_config) {
2443 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2448 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2450 struct ofproto *p = ofconn_get_ofproto(ofconn);
2451 struct ofputil_port_mod pm;
2452 struct ofport *port;
2455 error = reject_slave_controller(ofconn);
2460 error = ofputil_decode_port_mod(oh, &pm);
2465 port = ofproto_get_port(p, pm.port_no);
2467 return OFPERR_OFPPMFC_BAD_PORT;
2468 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2469 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2471 update_port_config(port, pm.config, pm.mask);
2473 netdev_set_advertisements(port->netdev, pm.advertise);
2480 handle_desc_stats_request(struct ofconn *ofconn,
2481 const struct ofp_header *request)
2483 static const char *default_mfr_desc = "Nicira, Inc.";
2484 static const char *default_hw_desc = "Open vSwitch";
2485 static const char *default_sw_desc = VERSION;
2486 static const char *default_serial_desc = "None";
2487 static const char *default_dp_desc = "None";
2489 struct ofproto *p = ofconn_get_ofproto(ofconn);
2490 struct ofp_desc_stats *ods;
2493 msg = ofpraw_alloc_stats_reply(request, 0);
2494 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2495 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2496 sizeof ods->mfr_desc);
2497 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2498 sizeof ods->hw_desc);
2499 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2500 sizeof ods->sw_desc);
2501 ovs_strlcpy(ods->serial_num,
2502 p->serial_desc ? p->serial_desc : default_serial_desc,
2503 sizeof ods->serial_num);
2504 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2505 sizeof ods->dp_desc);
2506 ofconn_send_reply(ofconn, msg);
2512 handle_table_stats_request(struct ofconn *ofconn,
2513 const struct ofp_header *request)
2515 struct ofproto *p = ofconn_get_ofproto(ofconn);
2516 struct ofp12_table_stats *ots;
2521 /* Set up default values.
2523 * ofp12_table_stats is used as a generic structure as
2524 * it is able to hold all the fields for ofp10_table_stats
2525 * and ofp11_table_stats (and of course itself).
2527 ots = xcalloc(p->n_tables, sizeof *ots);
2528 for (i = 0; i < p->n_tables; i++) {
2529 ots[i].table_id = i;
2530 sprintf(ots[i].name, "table%zu", i);
2531 ots[i].match = htonll(OFPXMT12_MASK);
2532 ots[i].wildcards = htonll(OFPXMT12_MASK);
2533 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2534 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2535 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2536 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2537 ots[i].metadata_match = htonll(UINT64_MAX);
2538 ots[i].metadata_write = htonll(UINT64_MAX);
2539 ots[i].instructions = htonl(OFPIT11_ALL);
2540 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2541 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2542 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2545 p->ofproto_class->get_tables(p, ots);
2547 /* Post-process the tables, dropping hidden tables. */
2548 n_tables = p->n_tables;
2549 for (i = 0; i < p->n_tables; i++) {
2550 const struct oftable *table = &p->tables[i];
2552 if (table->flags & OFTABLE_HIDDEN) {
2558 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2561 if (table->max_flows < ntohl(ots[i].max_entries)) {
2562 ots[i].max_entries = htonl(table->max_flows);
2566 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2567 ofconn_send_reply(ofconn, msg);
2575 append_port_stat(struct ofport *port, struct list *replies)
2577 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2579 calc_duration(port->created, time_msec(),
2580 &ops.duration_sec, &ops.duration_nsec);
2582 /* Intentionally ignore return value, since errors will set
2583 * 'stats' to all-1s, which is correct for OpenFlow, and
2584 * netdev_get_stats() will log errors. */
2585 ofproto_port_get_stats(port, &ops.stats);
2587 ofputil_append_port_stat(replies, &ops);
2591 handle_port_stats_request(struct ofconn *ofconn,
2592 const struct ofp_header *request)
2594 struct ofproto *p = ofconn_get_ofproto(ofconn);
2595 struct ofport *port;
2596 struct list replies;
2600 error = ofputil_decode_port_stats_request(request, &port_no);
2605 ofpmp_init(&replies, request);
2606 if (port_no != OFPP_ANY) {
2607 port = ofproto_get_port(p, port_no);
2609 append_port_stat(port, &replies);
2612 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2613 append_port_stat(port, &replies);
2617 ofconn_send_replies(ofconn, &replies);
2622 handle_port_desc_stats_request(struct ofconn *ofconn,
2623 const struct ofp_header *request)
2625 struct ofproto *p = ofconn_get_ofproto(ofconn);
2626 enum ofp_version version;
2627 struct ofport *port;
2628 struct list replies;
2630 ofpmp_init(&replies, request);
2632 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2633 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2634 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2637 ofconn_send_replies(ofconn, &replies);
2642 hash_cookie(ovs_be64 cookie)
2644 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
2645 (OVS_FORCE uint64_t)cookie);
2649 cookies_insert(struct ofproto *ofproto, struct rule *rule)
2651 hindex_insert(&ofproto->cookies, &rule->cookie_node,
2652 hash_cookie(rule->flow_cookie));
2656 cookies_remove(struct ofproto *ofproto, struct rule *rule)
2658 hindex_remove(&ofproto->cookies, &rule->cookie_node);
2662 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
2663 ovs_be64 new_cookie)
2665 if (new_cookie != rule->flow_cookie) {
2666 cookies_remove(ofproto, rule);
2668 rule->flow_cookie = new_cookie;
2670 cookies_insert(ofproto, rule);
2675 calc_duration(long long int start, long long int now,
2676 uint32_t *sec, uint32_t *nsec)
2678 long long int msecs = now - start;
2679 *sec = msecs / 1000;
2680 *nsec = (msecs % 1000) * (1000 * 1000);
2683 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2684 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2686 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2688 return (table_id == 0xff || table_id < ofproto->n_tables
2690 : OFPERR_OFPBRC_BAD_TABLE_ID);
2694 static struct oftable *
2695 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2697 struct oftable *table;
2699 for (table = &ofproto->tables[table_id];
2700 table < &ofproto->tables[ofproto->n_tables];
2702 if (!(table->flags & OFTABLE_HIDDEN)) {
2710 static struct oftable *
2711 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2713 if (table_id == 0xff) {
2714 return next_visible_table(ofproto, 0);
2715 } else if (table_id < ofproto->n_tables) {
2716 return &ofproto->tables[table_id];
2722 static struct oftable *
2723 next_matching_table(const struct ofproto *ofproto,
2724 const struct oftable *table, uint8_t table_id)
2726 return (table_id == 0xff
2727 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2731 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2733 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2734 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2736 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2737 * only once, for that table. (This can be used to access tables marked
2740 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2741 * entered at all. (Perhaps you should have validated TABLE_ID with
2742 * check_table_id().)
2744 * All parameters are evaluated multiple times.
2746 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2747 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2749 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2751 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2752 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2753 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2756 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2757 * to 'out_port' are included.
2759 * Hidden rules are always omitted.
2761 * Returns 0 on success, otherwise an OpenFlow error code. */
2763 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2764 const struct match *match,
2765 ovs_be64 cookie, ovs_be64 cookie_mask,
2766 ofp_port_t out_port, struct list *rules)
2768 struct oftable *table;
2772 error = check_table_id(ofproto, table_id);
2778 cls_rule_init(&cr, match, 0);
2780 if (cookie_mask == htonll(UINT64_MAX)) {
2783 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2784 &ofproto->cookies) {
2785 if (table_id != rule->table_id && table_id != 0xff) {
2788 if (ofproto_rule_is_hidden(rule)) {
2791 if (cls_rule_is_loose_match(&rule->cr, &cr.match)) {
2792 if (rule->pending) {
2793 error = OFPROTO_POSTPONE;
2796 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2797 && ofproto_rule_has_out_port(rule, out_port)) {
2798 list_push_back(rules, &rule->ofproto_node);
2805 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2806 struct cls_cursor cursor;
2809 cls_cursor_init(&cursor, &table->cls, &cr);
2810 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2811 if (rule->pending) {
2812 error = OFPROTO_POSTPONE;
2815 if (!ofproto_rule_is_hidden(rule)
2816 && ofproto_rule_has_out_port(rule, out_port)
2817 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2818 list_push_back(rules, &rule->ofproto_node);
2824 cls_rule_destroy(&cr);
2828 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2829 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2830 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2833 * If 'out_port' is anything other than OFPP_ANY, then only rules that output
2834 * to 'out_port' are included.
2836 * Hidden rules are always omitted.
2838 * Returns 0 on success, otherwise an OpenFlow error code. */
2840 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2841 const struct match *match, unsigned int priority,
2842 ovs_be64 cookie, ovs_be64 cookie_mask,
2843 ofp_port_t out_port, struct list *rules)
2845 struct oftable *table;
2849 error = check_table_id(ofproto, table_id);
2855 cls_rule_init(&cr, match, priority);
2857 if (cookie_mask == htonll(UINT64_MAX)) {
2860 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node, hash_cookie(cookie),
2861 &ofproto->cookies) {
2862 if (table_id != rule->table_id && table_id != 0xff) {
2865 if (ofproto_rule_is_hidden(rule)) {
2868 if (cls_rule_equal(&rule->cr, &cr)) {
2869 if (rule->pending) {
2870 error = OFPROTO_POSTPONE;
2873 if (rule->flow_cookie == cookie /* Hash collisions possible. */
2874 && ofproto_rule_has_out_port(rule, out_port)) {
2875 list_push_back(rules, &rule->ofproto_node);
2882 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2885 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2888 if (rule->pending) {
2889 error = OFPROTO_POSTPONE;
2892 if (!ofproto_rule_is_hidden(rule)
2893 && ofproto_rule_has_out_port(rule, out_port)
2894 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2895 list_push_back(rules, &rule->ofproto_node);
2901 cls_rule_destroy(&cr);
2905 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2906 * forced into the range of a uint16_t. */
2908 age_secs(long long int age_ms)
2910 return (age_ms < 0 ? 0
2911 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2912 : (unsigned int) age_ms / 1000);
2916 handle_flow_stats_request(struct ofconn *ofconn,
2917 const struct ofp_header *request)
2919 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2920 struct ofputil_flow_stats_request fsr;
2921 struct list replies;
2926 error = ofputil_decode_flow_stats_request(&fsr, request);
2931 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2932 fsr.cookie, fsr.cookie_mask,
2933 fsr.out_port, &rules);
2938 ofpmp_init(&replies, request);
2939 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2940 long long int now = time_msec();
2941 struct ofputil_flow_stats fs;
2943 minimatch_expand(&rule->cr.match, &fs.match);
2944 fs.priority = rule->cr.priority;
2945 fs.cookie = rule->flow_cookie;
2946 fs.table_id = rule->table_id;
2947 calc_duration(rule->created, now, &fs.duration_sec, &fs.duration_nsec);
2948 fs.idle_timeout = rule->idle_timeout;
2949 fs.hard_timeout = rule->hard_timeout;
2950 fs.idle_age = age_secs(now - rule->used);
2951 fs.hard_age = age_secs(now - rule->modified);
2952 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2954 fs.ofpacts = rule->ofpacts;
2955 fs.ofpacts_len = rule->ofpacts_len;
2957 if (rule->send_flow_removed) {
2958 fs.flags |= OFPFF_SEND_FLOW_REM;
2959 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
2960 and OFPFF13_NO_BYT_COUNTS */
2962 ofputil_append_flow_stats_reply(&fs, &replies);
2964 ofconn_send_replies(ofconn, &replies);
2970 flow_stats_ds(struct rule *rule, struct ds *results)
2972 uint64_t packet_count, byte_count;
2974 rule->ofproto->ofproto_class->rule_get_stats(rule,
2975 &packet_count, &byte_count);
2977 if (rule->table_id != 0) {
2978 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2980 ds_put_format(results, "duration=%llds, ",
2981 (time_msec() - rule->created) / 1000);
2982 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2983 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2984 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2985 cls_rule_format(&rule->cr, results);
2986 ds_put_char(results, ',');
2987 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2988 ds_put_cstr(results, "\n");
2991 /* Adds a pretty-printed description of all flows to 'results', including
2992 * hidden flows (e.g., set up by in-band control). */
2994 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2996 struct oftable *table;
2998 OFPROTO_FOR_EACH_TABLE (table, p) {
2999 struct cls_cursor cursor;
3002 cls_cursor_init(&cursor, &table->cls, NULL);
3003 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3004 flow_stats_ds(rule, results);
3009 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3010 * '*engine_type' and '*engine_id', respectively. */
3012 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3013 uint8_t *engine_type, uint8_t *engine_id)
3015 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3018 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3019 * true if the port's CFM status was successfully stored into '*status'.
3020 * Returns false if the port did not have CFM configured, in which case
3021 * '*status' is indeterminate.
3023 * The caller must provide and owns '*status', but it does not own and must not
3024 * modify or free the array returned in 'status->rmps'. */
3026 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3027 struct ofproto_cfm_status *status)
3029 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3031 && ofproto->ofproto_class->get_cfm_status
3032 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3036 handle_aggregate_stats_request(struct ofconn *ofconn,
3037 const struct ofp_header *oh)
3039 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3040 struct ofputil_flow_stats_request request;
3041 struct ofputil_aggregate_stats stats;
3042 bool unknown_packets, unknown_bytes;
3043 struct ofpbuf *reply;
3048 error = ofputil_decode_flow_stats_request(&request, oh);
3053 error = collect_rules_loose(ofproto, request.table_id, &request.match,
3054 request.cookie, request.cookie_mask,
3055 request.out_port, &rules);
3060 memset(&stats, 0, sizeof stats);
3061 unknown_packets = unknown_bytes = false;
3062 LIST_FOR_EACH (rule, ofproto_node, &rules) {
3063 uint64_t packet_count;
3064 uint64_t byte_count;
3066 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3069 if (packet_count == UINT64_MAX) {
3070 unknown_packets = true;
3072 stats.packet_count += packet_count;
3075 if (byte_count == UINT64_MAX) {
3076 unknown_bytes = true;
3078 stats.byte_count += byte_count;
3083 if (unknown_packets) {
3084 stats.packet_count = UINT64_MAX;
3086 if (unknown_bytes) {
3087 stats.byte_count = UINT64_MAX;
3090 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3091 ofconn_send_reply(ofconn, reply);
3096 struct queue_stats_cbdata {
3097 struct ofport *ofport;
3098 struct list replies;
3102 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3103 const struct netdev_queue_stats *stats)
3106 struct ofputil_queue_stats oqs = {
3107 .port_no = cbdata->ofport->pp.port_no,
3108 .queue_id = queue_id,
3111 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3115 handle_queue_stats_dump_cb(uint32_t queue_id,
3116 struct netdev_queue_stats *stats,
3119 struct queue_stats_cbdata *cbdata = cbdata_;
3121 put_queue_stats(cbdata, queue_id, stats);
3125 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3126 struct queue_stats_cbdata *cbdata)
3128 cbdata->ofport = port;
3129 if (queue_id == OFPQ_ALL) {
3130 netdev_dump_queue_stats(port->netdev,
3131 handle_queue_stats_dump_cb, cbdata);
3133 struct netdev_queue_stats stats;
3135 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3136 put_queue_stats(cbdata, queue_id, &stats);
3138 return OFPERR_OFPQOFC_BAD_QUEUE;
3145 handle_queue_stats_request(struct ofconn *ofconn,
3146 const struct ofp_header *rq)
3148 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3149 struct queue_stats_cbdata cbdata;
3150 struct ofport *port;
3152 struct ofputil_queue_stats_request oqsr;
3154 COVERAGE_INC(ofproto_queue_req);
3156 ofpmp_init(&cbdata.replies, rq);
3158 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3163 if (oqsr.port_no == OFPP_ANY) {
3164 error = OFPERR_OFPQOFC_BAD_QUEUE;
3165 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3166 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3171 port = ofproto_get_port(ofproto, oqsr.port_no);
3173 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3174 : OFPERR_OFPQOFC_BAD_PORT);
3177 ofconn_send_replies(ofconn, &cbdata.replies);
3179 ofpbuf_list_delete(&cbdata.replies);
3186 is_flow_deletion_pending(const struct ofproto *ofproto,
3187 const struct cls_rule *cls_rule,
3190 if (!hmap_is_empty(&ofproto->deletions)) {
3191 struct ofoperation *op;
3193 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3194 cls_rule_hash(cls_rule, table_id),
3195 &ofproto->deletions) {
3196 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3205 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3206 * in which no matching flow already exists in the flow table.
3208 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3209 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3210 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3211 * initiated now but may be retried later.
3213 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
3214 * ownership remains with the caller.
3216 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3219 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3220 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
3222 struct oftable *table;
3223 struct ofopgroup *group;
3224 struct rule *victim;
3229 error = check_table_id(ofproto, fm->table_id);
3235 if (fm->table_id == 0xff) {
3236 if (ofproto->ofproto_class->rule_choose_table) {
3237 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3243 ovs_assert(table_id < ofproto->n_tables);
3247 } else if (fm->table_id < ofproto->n_tables) {
3248 table_id = fm->table_id;
3250 return OFPERR_OFPBRC_BAD_TABLE_ID;
3253 table = &ofproto->tables[table_id];
3255 if (table->flags & OFTABLE_READONLY) {
3256 return OFPERR_OFPBRC_EPERM;
3259 /* Allocate new rule and initialize classifier rule. */
3260 rule = ofproto->ofproto_class->rule_alloc();
3262 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3263 ofproto->name, strerror(error));
3266 cls_rule_init(&rule->cr, &fm->match, fm->priority);
3268 /* Serialize against pending deletion. */
3269 if (is_flow_deletion_pending(ofproto, &rule->cr, table_id)) {
3270 cls_rule_destroy(&rule->cr);
3271 ofproto->ofproto_class->rule_dealloc(rule);
3272 return OFPROTO_POSTPONE;
3275 /* Check for overlap, if requested. */
3276 if (fm->flags & OFPFF_CHECK_OVERLAP
3277 && classifier_rule_overlaps(&table->cls, &rule->cr)) {
3278 cls_rule_destroy(&rule->cr);
3279 ofproto->ofproto_class->rule_dealloc(rule);
3280 return OFPERR_OFPFMFC_OVERLAP;
3283 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3285 rule->ofproto = ofproto;
3286 rule->pending = NULL;
3287 rule->flow_cookie = fm->new_cookie;
3288 rule->created = rule->modified = rule->used = time_msec();
3289 rule->idle_timeout = fm->idle_timeout;
3290 rule->hard_timeout = fm->hard_timeout;
3291 rule->table_id = table - ofproto->tables;
3292 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
3293 /* FIXME: Implement OF 1.3 flags OFPFF13_NO_PKT_COUNTS
3294 and OFPFF13_NO_BYT_COUNTS */
3295 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3296 rule->ofpacts_len = fm->ofpacts_len;
3297 rule->evictable = true;
3298 rule->eviction_group = NULL;
3299 list_init(&rule->expirable);
3300 rule->monitor_flags = 0;
3301 rule->add_seqno = 0;
3302 rule->modify_seqno = 0;
3304 /* Insert new rule. */
3305 victim = oftable_replace_rule(rule);
3306 if (victim && !rule_is_modifiable(victim)) {
3307 error = OFPERR_OFPBRC_EPERM;
3308 } else if (victim && victim->pending) {
3309 error = OFPROTO_POSTPONE;
3311 struct ofoperation *op;
3314 if (classifier_count(&table->cls) > table->max_flows) {
3317 was_evictable = rule->evictable;
3318 rule->evictable = false;
3319 evict = choose_rule_to_evict(table);
3320 rule->evictable = was_evictable;
3323 error = OFPERR_OFPFMFC_TABLE_FULL;
3325 } else if (evict->pending) {
3326 error = OFPROTO_POSTPONE;
3333 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3334 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3335 op->victim = victim;
3337 error = ofproto->ofproto_class->rule_construct(rule);
3339 op->group->n_running--;
3340 ofoperation_destroy(rule->pending);
3342 delete_flow__(evict, group);
3344 ofopgroup_submit(group);
3348 /* Back out if an error occurred. */
3350 oftable_substitute_rule(rule, victim);
3351 ofproto_rule_destroy__(rule);
3356 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3358 /* Modifies the rules listed in 'rules', changing their actions to match those
3361 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3364 * Returns 0 on success, otherwise an OpenFlow error code. */
3366 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3367 const struct ofputil_flow_mod *fm,
3368 const struct ofp_header *request, struct list *rules)
3370 struct ofopgroup *group;
3374 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3375 error = OFPERR_OFPBRC_EPERM;
3376 LIST_FOR_EACH (rule, ofproto_node, rules) {
3377 struct ofoperation *op;
3378 bool actions_changed;
3380 /* FIXME: Implement OFPFF12_RESET_COUNTS */
3382 if (rule_is_modifiable(rule)) {
3383 /* At least one rule is modifiable, don't report EPERM error. */
3389 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3390 rule->ofpacts, rule->ofpacts_len);
3392 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3394 if (fm->new_cookie != htonll(UINT64_MAX)) {
3395 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3397 if (actions_changed) {
3398 op->ofpacts = rule->ofpacts;
3399 op->ofpacts_len = rule->ofpacts_len;
3400 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3401 rule->ofpacts_len = fm->ofpacts_len;
3402 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3404 ofoperation_complete(op, 0);
3407 ofopgroup_submit(group);
3413 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
3414 const struct ofputil_flow_mod *fm,
3415 const struct ofp_header *request)
3417 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
3420 return add_flow(ofproto, ofconn, fm, request);
3423 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3426 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3429 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3430 const struct ofputil_flow_mod *fm,
3431 const struct ofp_header *request)
3436 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3437 fm->cookie, fm->cookie_mask,
3441 } else if (list_is_empty(&rules)) {
3442 return modify_flows_add(ofproto, ofconn, fm, request);
3444 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3448 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3451 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3454 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3455 const struct ofputil_flow_mod *fm,
3456 const struct ofp_header *request)
3461 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3462 fm->priority, fm->cookie, fm->cookie_mask,
3467 } else if (list_is_empty(&rules)) {
3468 return modify_flows_add(ofproto, ofconn, fm, request);
3470 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3471 fm, request, &rules)
3476 /* OFPFC_DELETE implementation. */
3479 delete_flow__(struct rule *rule, struct ofopgroup *group)
3481 struct ofproto *ofproto = rule->ofproto;
3483 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3485 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3486 oftable_remove_rule(rule);
3487 ofproto->ofproto_class->rule_destruct(rule);
3490 /* Deletes the rules listed in 'rules'.
3492 * Returns 0 on success, otherwise an OpenFlow error code. */
3494 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3495 const struct ofp_header *request, struct list *rules)
3497 struct rule *rule, *next;
3498 struct ofopgroup *group;
3500 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3501 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3502 delete_flow__(rule, group);
3504 ofopgroup_submit(group);
3509 /* Implements OFPFC_DELETE. */
3511 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3512 const struct ofputil_flow_mod *fm,
3513 const struct ofp_header *request)
3518 error = collect_rules_loose(ofproto, fm->table_id, &fm->match,
3519 fm->cookie, fm->cookie_mask,
3520 fm->out_port, &rules);
3521 return (error ? error
3522 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3527 /* Implements OFPFC_DELETE_STRICT. */
3529 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3530 const struct ofputil_flow_mod *fm,
3531 const struct ofp_header *request)
3536 error = collect_rules_strict(ofproto, fm->table_id, &fm->match,
3537 fm->priority, fm->cookie, fm->cookie_mask,
3538 fm->out_port, &rules);
3539 return (error ? error
3540 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3546 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3548 struct ofputil_flow_removed fr;
3550 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3554 minimatch_expand(&rule->cr.match, &fr.match);
3555 fr.priority = rule->cr.priority;
3556 fr.cookie = rule->flow_cookie;
3558 fr.table_id = rule->table_id;
3559 calc_duration(rule->created, time_msec(),
3560 &fr.duration_sec, &fr.duration_nsec);
3561 fr.idle_timeout = rule->idle_timeout;
3562 fr.hard_timeout = rule->hard_timeout;
3563 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3566 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3570 ofproto_rule_update_used(struct rule *rule, long long int used)
3572 if (used > rule->used) {
3573 struct eviction_group *evg = rule->eviction_group;
3577 heap_change(&evg->rules, &rule->evg_node,
3578 rule_eviction_priority(rule));
3583 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3584 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3587 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3590 * ofproto implementation ->run() functions should use this function to expire
3591 * OpenFlow flows. */
3593 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3595 struct ofproto *ofproto = rule->ofproto;
3596 struct ofopgroup *group;
3598 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3600 ofproto_rule_send_removed(rule, reason);
3602 group = ofopgroup_create_unattached(ofproto);
3603 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3604 oftable_remove_rule(rule);
3605 ofproto->ofproto_class->rule_destruct(rule);
3606 ofopgroup_submit(group);
3610 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3612 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3613 struct ofputil_flow_mod fm;
3614 uint64_t ofpacts_stub[1024 / 8];
3615 struct ofpbuf ofpacts;
3619 error = reject_slave_controller(ofconn);
3624 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3625 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3628 error = ofpacts_check(fm.ofpacts, fm.ofpacts_len,
3629 &fm.match.flow, ofproto->max_ports);
3632 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
3635 goto exit_free_ofpacts;
3638 /* Record the operation for logging a summary report. */
3639 switch (fm.command) {
3645 case OFPFC_MODIFY_STRICT:
3646 ofproto->n_modify++;
3650 case OFPFC_DELETE_STRICT:
3651 ofproto->n_delete++;
3656 if (ofproto->next_op_report == LLONG_MAX) {
3657 ofproto->first_op = now;
3658 ofproto->next_op_report = MAX(now + 10 * 1000,
3659 ofproto->op_backoff);
3660 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3662 ofproto->last_op = now;
3665 ofpbuf_uninit(&ofpacts);
3671 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3672 const struct ofputil_flow_mod *fm,
3673 const struct ofp_header *oh)
3675 if (ofproto->n_pending >= 50) {
3676 ovs_assert(!list_is_empty(&ofproto->pending));
3677 return OFPROTO_POSTPONE;
3680 switch (fm->command) {
3682 return add_flow(ofproto, ofconn, fm, oh);
3685 return modify_flows_loose(ofproto, ofconn, fm, oh);
3687 case OFPFC_MODIFY_STRICT:
3688 return modify_flow_strict(ofproto, ofconn, fm, oh);
3691 return delete_flows_loose(ofproto, ofconn, fm, oh);
3693 case OFPFC_DELETE_STRICT:
3694 return delete_flow_strict(ofproto, ofconn, fm, oh);
3697 if (fm->command > 0xff) {
3698 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3699 "flow_mod_table_id extension is not enabled",
3702 return OFPERR_OFPFMFC_BAD_COMMAND;
3707 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3709 struct ofputil_role_request request;
3710 struct ofputil_role_request reply;
3714 error = ofputil_decode_role_message(oh, &request);
3719 if (request.role != OFPCR12_ROLE_NOCHANGE) {
3720 if (ofconn_get_role(ofconn) != request.role
3721 && ofconn_has_pending_opgroups(ofconn)) {
3722 return OFPROTO_POSTPONE;
3725 if (request.have_generation_id
3726 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
3727 return OFPERR_OFPRRFC_STALE;
3730 ofconn_set_role(ofconn, request.role);
3733 reply.role = ofconn_get_role(ofconn);
3734 reply.have_generation_id = ofconn_get_master_election_id(
3735 ofconn, &reply.generation_id);
3736 buf = ofputil_encode_role_reply(oh, &reply);
3737 ofconn_send_reply(ofconn, buf);
3743 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3744 const struct ofp_header *oh)
3746 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3747 enum ofputil_protocol cur, next;
3749 cur = ofconn_get_protocol(ofconn);
3750 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3751 ofconn_set_protocol(ofconn, next);
3757 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3759 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3760 enum ofputil_protocol cur, next;
3761 enum ofputil_protocol next_base;
3763 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3765 return OFPERR_OFPBRC_EPERM;
3768 cur = ofconn_get_protocol(ofconn);
3769 next = ofputil_protocol_set_base(cur, next_base);
3770 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3771 /* Avoid sending async messages in surprising protocol. */
3772 return OFPROTO_POSTPONE;
3775 ofconn_set_protocol(ofconn, next);
3780 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3781 const struct ofp_header *oh)
3783 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3786 format = ntohl(msg->format);
3787 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3788 return OFPERR_OFPBRC_EPERM;
3791 if (format != ofconn_get_packet_in_format(ofconn)
3792 && ofconn_has_pending_opgroups(ofconn)) {
3793 /* Avoid sending async message in surprsing packet in format. */
3794 return OFPROTO_POSTPONE;
3797 ofconn_set_packet_in_format(ofconn, format);
3802 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3804 const struct nx_async_config *msg = ofpmsg_body(oh);
3805 uint32_t master[OAM_N_TYPES];
3806 uint32_t slave[OAM_N_TYPES];
3808 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3809 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3810 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3812 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3813 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3814 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3816 ofconn_set_async_config(ofconn, master, slave);
3817 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3818 !ofconn_get_miss_send_len(ofconn)) {
3819 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3826 handle_nxt_set_controller_id(struct ofconn *ofconn,
3827 const struct ofp_header *oh)
3829 const struct nx_controller_id *nci = ofpmsg_body(oh);
3831 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3832 return OFPERR_NXBRC_MUST_BE_ZERO;
3835 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3840 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3844 if (ofconn_has_pending_opgroups(ofconn)) {
3845 return OFPROTO_POSTPONE;
3848 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3849 ? OFPRAW_OFPT10_BARRIER_REPLY
3850 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3851 ofconn_send_reply(ofconn, buf);
3856 ofproto_compose_flow_refresh_update(const struct rule *rule,
3857 enum nx_flow_monitor_flags flags,
3860 struct ofoperation *op = rule->pending;
3861 struct ofputil_flow_update fu;
3864 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3865 /* We'll report the final flow when the operation completes. Reporting
3866 * it now would cause a duplicate report later. */
3870 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3871 ? NXFME_ADDED : NXFME_MODIFIED);
3873 fu.idle_timeout = rule->idle_timeout;
3874 fu.hard_timeout = rule->hard_timeout;
3875 fu.table_id = rule->table_id;
3876 fu.cookie = rule->flow_cookie;
3877 minimatch_expand(&rule->cr.match, &match);
3879 fu.priority = rule->cr.priority;
3880 if (!(flags & NXFMF_ACTIONS)) {
3884 fu.ofpacts = rule->ofpacts;
3885 fu.ofpacts_len = rule->ofpacts_len;
3887 /* An operation is in progress. Use the previous version of the flow's
3888 * actions, so that when the operation commits we report the change. */
3890 case OFOPERATION_ADD:
3891 /* We already verified that there was a victim. */
3892 fu.ofpacts = op->victim->ofpacts;
3893 fu.ofpacts_len = op->victim->ofpacts_len;
3896 case OFOPERATION_MODIFY:
3898 fu.ofpacts = op->ofpacts;
3899 fu.ofpacts_len = op->ofpacts_len;
3901 fu.ofpacts = rule->ofpacts;
3902 fu.ofpacts_len = rule->ofpacts_len;
3906 case OFOPERATION_DELETE:
3907 fu.ofpacts = rule->ofpacts;
3908 fu.ofpacts_len = rule->ofpacts_len;
3916 if (list_is_empty(msgs)) {
3917 ofputil_start_flow_update(msgs);
3919 ofputil_append_flow_update(&fu, msgs);
3923 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3927 LIST_FOR_EACH (rule, ofproto_node, rules) {
3928 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3929 rule->monitor_flags = 0;
3931 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3936 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3937 struct rule *rule, uint64_t seqno,
3940 enum nx_flow_monitor_flags update;
3942 if (ofproto_rule_is_hidden(rule)) {
3947 ? ofoperation_has_out_port(rule->pending, m->out_port)
3948 : ofproto_rule_has_out_port(rule, m->out_port))) {
3953 if (rule->add_seqno > seqno) {
3954 update = NXFMF_ADD | NXFMF_MODIFY;
3955 } else if (rule->modify_seqno > seqno) {
3956 update = NXFMF_MODIFY;
3961 if (!(m->flags & update)) {
3965 update = NXFMF_INITIAL;
3968 if (!rule->monitor_flags) {
3969 list_push_back(rules, &rule->ofproto_node);
3971 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3975 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3979 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3980 const struct ofoperation *op;
3981 const struct oftable *table;
3982 struct cls_rule target;
3984 cls_rule_init_from_minimatch(&target, &m->match, 0);
3985 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3986 struct cls_cursor cursor;
3989 cls_cursor_init(&cursor, &table->cls, &target);
3990 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3991 ovs_assert(!rule->pending); /* XXX */
3992 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3996 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3997 struct rule *rule = op->rule;
3999 if (((m->table_id == 0xff
4000 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4001 : m->table_id == rule->table_id))
4002 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4003 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4006 cls_rule_destroy(&target);
4010 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4013 if (m->flags & NXFMF_INITIAL) {
4014 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4019 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4020 uint64_t seqno, struct list *rules)
4022 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4026 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4028 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4029 struct ofmonitor **monitors;
4030 size_t n_monitors, allocated_monitors;
4031 struct list replies;
4038 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4040 n_monitors = allocated_monitors = 0;
4042 struct ofputil_flow_monitor_request request;
4043 struct ofmonitor *m;
4046 retval = ofputil_decode_flow_monitor_request(&request, &b);
4047 if (retval == EOF) {
4049 } else if (retval) {
4054 if (request.table_id != 0xff
4055 && request.table_id >= ofproto->n_tables) {
4056 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4060 error = ofmonitor_create(&request, ofconn, &m);
4065 if (n_monitors >= allocated_monitors) {
4066 monitors = x2nrealloc(monitors, &allocated_monitors,
4069 monitors[n_monitors++] = m;
4073 for (i = 0; i < n_monitors; i++) {
4074 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4077 ofpmp_init(&replies, oh);
4078 ofmonitor_compose_refresh_updates(&rules, &replies);
4079 ofconn_send_replies(ofconn, &replies);
4086 for (i = 0; i < n_monitors; i++) {
4087 ofmonitor_destroy(monitors[i]);
4094 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4096 struct ofmonitor *m;
4099 id = ofputil_decode_flow_monitor_cancel(oh);
4100 m = ofmonitor_lookup(ofconn, id);
4102 return OFPERR_NXBRC_FM_BAD_ID;
4105 ofmonitor_destroy(m);
4110 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
4112 const struct ofp_header *oh = msg->data;
4116 error = ofptype_decode(&type, oh);
4122 /* OpenFlow requests. */
4123 case OFPTYPE_ECHO_REQUEST:
4124 return handle_echo_request(ofconn, oh);
4126 case OFPTYPE_FEATURES_REQUEST:
4127 return handle_features_request(ofconn, oh);
4129 case OFPTYPE_GET_CONFIG_REQUEST:
4130 return handle_get_config_request(ofconn, oh);
4132 case OFPTYPE_SET_CONFIG:
4133 return handle_set_config(ofconn, oh);
4135 case OFPTYPE_PACKET_OUT:
4136 return handle_packet_out(ofconn, oh);
4138 case OFPTYPE_PORT_MOD:
4139 return handle_port_mod(ofconn, oh);
4141 case OFPTYPE_FLOW_MOD:
4142 return handle_flow_mod(ofconn, oh);
4144 case OFPTYPE_BARRIER_REQUEST:
4145 return handle_barrier_request(ofconn, oh);
4147 case OFPTYPE_ROLE_REQUEST:
4148 return handle_role_request(ofconn, oh);
4150 /* OpenFlow replies. */
4151 case OFPTYPE_ECHO_REPLY:
4154 /* Nicira extension requests. */
4155 case OFPTYPE_FLOW_MOD_TABLE_ID:
4156 return handle_nxt_flow_mod_table_id(ofconn, oh);
4158 case OFPTYPE_SET_FLOW_FORMAT:
4159 return handle_nxt_set_flow_format(ofconn, oh);
4161 case OFPTYPE_SET_PACKET_IN_FORMAT:
4162 return handle_nxt_set_packet_in_format(ofconn, oh);
4164 case OFPTYPE_SET_CONTROLLER_ID:
4165 return handle_nxt_set_controller_id(ofconn, oh);
4167 case OFPTYPE_FLOW_AGE:
4168 /* Nothing to do. */
4171 case OFPTYPE_FLOW_MONITOR_CANCEL:
4172 return handle_flow_monitor_cancel(ofconn, oh);
4174 case OFPTYPE_SET_ASYNC_CONFIG:
4175 return handle_nxt_set_async_config(ofconn, oh);
4177 /* Statistics requests. */
4178 case OFPTYPE_DESC_STATS_REQUEST:
4179 return handle_desc_stats_request(ofconn, oh);
4181 case OFPTYPE_FLOW_STATS_REQUEST:
4182 return handle_flow_stats_request(ofconn, oh);
4184 case OFPTYPE_AGGREGATE_STATS_REQUEST:
4185 return handle_aggregate_stats_request(ofconn, oh);
4187 case OFPTYPE_TABLE_STATS_REQUEST:
4188 return handle_table_stats_request(ofconn, oh);
4190 case OFPTYPE_PORT_STATS_REQUEST:
4191 return handle_port_stats_request(ofconn, oh);
4193 case OFPTYPE_QUEUE_STATS_REQUEST:
4194 return handle_queue_stats_request(ofconn, oh);
4196 case OFPTYPE_PORT_DESC_STATS_REQUEST:
4197 return handle_port_desc_stats_request(ofconn, oh);
4199 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
4200 return handle_flow_monitor_request(ofconn, oh);
4202 /* FIXME: Change the following once they are implemented: */
4203 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
4204 case OFPTYPE_GET_ASYNC_REQUEST:
4205 case OFPTYPE_METER_MOD:
4206 case OFPTYPE_GROUP_REQUEST:
4207 case OFPTYPE_GROUP_DESC_REQUEST:
4208 case OFPTYPE_GROUP_FEATURES_REQUEST:
4209 case OFPTYPE_METER_REQUEST:
4210 case OFPTYPE_METER_CONFIG_REQUEST:
4211 case OFPTYPE_METER_FEATURES_REQUEST:
4212 case OFPTYPE_TABLE_FEATURES_REQUEST:
4213 return OFPERR_OFPBRC_BAD_TYPE;
4217 case OFPTYPE_FEATURES_REPLY:
4218 case OFPTYPE_GET_CONFIG_REPLY:
4219 case OFPTYPE_PACKET_IN:
4220 case OFPTYPE_FLOW_REMOVED:
4221 case OFPTYPE_PORT_STATUS:
4222 case OFPTYPE_BARRIER_REPLY:
4223 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
4224 case OFPTYPE_DESC_STATS_REPLY:
4225 case OFPTYPE_FLOW_STATS_REPLY:
4226 case OFPTYPE_QUEUE_STATS_REPLY:
4227 case OFPTYPE_PORT_STATS_REPLY:
4228 case OFPTYPE_TABLE_STATS_REPLY:
4229 case OFPTYPE_AGGREGATE_STATS_REPLY:
4230 case OFPTYPE_PORT_DESC_STATS_REPLY:
4231 case OFPTYPE_ROLE_REPLY:
4232 case OFPTYPE_FLOW_MONITOR_PAUSED:
4233 case OFPTYPE_FLOW_MONITOR_RESUMED:
4234 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
4235 case OFPTYPE_GET_ASYNC_REPLY:
4236 case OFPTYPE_GROUP_REPLY:
4237 case OFPTYPE_GROUP_DESC_REPLY:
4238 case OFPTYPE_GROUP_FEATURES_REPLY:
4239 case OFPTYPE_METER_REPLY:
4240 case OFPTYPE_METER_CONFIG_REPLY:
4241 case OFPTYPE_METER_FEATURES_REPLY:
4242 case OFPTYPE_TABLE_FEATURES_REPLY:
4244 return OFPERR_OFPBRC_BAD_TYPE;
4249 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
4251 int error = handle_openflow__(ofconn, ofp_msg);
4252 if (error && error != OFPROTO_POSTPONE) {
4253 ofconn_send_error(ofconn, ofp_msg->data, error);
4255 COVERAGE_INC(ofproto_recv_openflow);
4256 return error != OFPROTO_POSTPONE;
4259 /* Asynchronous operations. */
4261 /* Creates and returns a new ofopgroup that is not associated with any
4262 * OpenFlow connection.
4264 * The caller should add operations to the returned group with
4265 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4266 static struct ofopgroup *
4267 ofopgroup_create_unattached(struct ofproto *ofproto)
4269 struct ofopgroup *group = xzalloc(sizeof *group);
4270 group->ofproto = ofproto;
4271 list_init(&group->ofproto_node);
4272 list_init(&group->ops);
4273 list_init(&group->ofconn_node);
4277 /* Creates and returns a new ofopgroup for 'ofproto'.
4279 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
4280 * connection. The 'request' and 'buffer_id' arguments are ignored.
4282 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
4283 * If the ofopgroup eventually fails, then the error reply will include
4284 * 'request'. If the ofopgroup eventually succeeds, then the packet with
4285 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
4287 * The caller should add operations to the returned group with
4288 * ofoperation_create() and then submit it with ofopgroup_submit(). */
4289 static struct ofopgroup *
4290 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
4291 const struct ofp_header *request, uint32_t buffer_id)
4293 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
4295 size_t request_len = ntohs(request->length);
4297 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
4299 ofconn_add_opgroup(ofconn, &group->ofconn_node);
4300 group->ofconn = ofconn;
4301 group->request = xmemdup(request, MIN(request_len, 64));
4302 group->buffer_id = buffer_id;
4307 /* Submits 'group' for processing.
4309 * If 'group' contains no operations (e.g. none were ever added, or all of the
4310 * ones that were added completed synchronously), then it is destroyed
4311 * immediately. Otherwise it is added to the ofproto's list of pending
4314 ofopgroup_submit(struct ofopgroup *group)
4316 if (!group->n_running) {
4317 ofopgroup_complete(group);
4319 list_push_back(&group->ofproto->pending, &group->ofproto_node);
4320 group->ofproto->n_pending++;
4325 ofopgroup_complete(struct ofopgroup *group)
4327 struct ofproto *ofproto = group->ofproto;
4329 struct ofconn *abbrev_ofconn;
4330 ovs_be32 abbrev_xid;
4332 struct ofoperation *op, *next_op;
4335 ovs_assert(!group->n_running);
4338 LIST_FOR_EACH (op, group_node, &group->ops) {
4345 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
4346 LIST_FOR_EACH (op, group_node, &group->ops) {
4347 if (op->type != OFOPERATION_DELETE) {
4348 struct ofpbuf *packet;
4351 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
4355 error = rule_execute(op->rule, in_port, packet);
4362 if (!error && !list_is_empty(&group->ofconn_node)) {
4363 abbrev_ofconn = group->ofconn;
4364 abbrev_xid = group->request->xid;
4366 abbrev_ofconn = NULL;
4367 abbrev_xid = htonl(0);
4369 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
4370 struct rule *rule = op->rule;
4372 /* We generally want to report the change to active OpenFlow flow
4373 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
4375 - The operation failed.
4377 - The affected rule is not visible to controllers.
4379 - The operation's only effect was to update rule->modified. */
4381 || ofproto_rule_is_hidden(rule)
4382 || (op->type == OFOPERATION_MODIFY
4384 && rule->flow_cookie == op->flow_cookie))) {
4385 /* Check that we can just cast from ofoperation_type to
4386 * nx_flow_update_event. */
4387 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
4389 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
4391 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
4394 ofmonitor_report(ofproto->connmgr, rule,
4395 (enum nx_flow_update_event) op->type,
4396 op->reason, abbrev_ofconn, abbrev_xid);
4399 rule->pending = NULL;
4402 case OFOPERATION_ADD:
4406 ofproto_rule_destroy__(op->victim);
4407 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
4408 if (vid_mask == VLAN_VID_MASK) {
4409 if (ofproto->vlan_bitmap) {
4410 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4411 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4412 bitmap_set1(ofproto->vlan_bitmap, vid);
4413 ofproto->vlans_changed = true;
4416 ofproto->vlans_changed = true;
4420 oftable_substitute_rule(rule, op->victim);
4421 ofproto_rule_destroy__(rule);
4425 case OFOPERATION_DELETE:
4426 ovs_assert(!op->error);
4427 ofproto_rule_destroy__(rule);
4431 case OFOPERATION_MODIFY:
4433 rule->modified = time_msec();
4435 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
4437 free(rule->ofpacts);
4438 rule->ofpacts = op->ofpacts;
4439 rule->ofpacts_len = op->ofpacts_len;
4441 op->ofpacts_len = 0;
4450 ofoperation_destroy(op);
4453 ofmonitor_flush(ofproto->connmgr);
4455 if (!list_is_empty(&group->ofproto_node)) {
4456 ovs_assert(ofproto->n_pending > 0);
4457 ofproto->n_pending--;
4458 list_remove(&group->ofproto_node);
4460 if (!list_is_empty(&group->ofconn_node)) {
4461 list_remove(&group->ofconn_node);
4463 ofconn_send_error(group->ofconn, group->request, error);
4465 connmgr_retry(ofproto->connmgr);
4467 free(group->request);
4471 /* Initiates a new operation on 'rule', of the specified 'type', within
4472 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4474 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4475 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4477 * Returns the newly created ofoperation (which is also available as
4478 * rule->pending). */
4479 static struct ofoperation *
4480 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4481 enum ofoperation_type type,
4482 enum ofp_flow_removed_reason reason)
4484 struct ofproto *ofproto = group->ofproto;
4485 struct ofoperation *op;
4487 ovs_assert(!rule->pending);
4489 op = rule->pending = xzalloc(sizeof *op);
4491 list_push_back(&group->ops, &op->group_node);
4494 op->reason = reason;
4495 op->flow_cookie = rule->flow_cookie;
4499 if (type == OFOPERATION_DELETE) {
4500 hmap_insert(&ofproto->deletions, &op->hmap_node,
4501 cls_rule_hash(&rule->cr, rule->table_id));
4508 ofoperation_destroy(struct ofoperation *op)
4510 struct ofopgroup *group = op->group;
4513 op->rule->pending = NULL;
4515 if (op->type == OFOPERATION_DELETE) {
4516 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4518 list_remove(&op->group_node);
4523 /* Indicates that 'op' completed with status 'error', which is either 0 to
4524 * indicate success or an OpenFlow error code on failure.
4526 * If 'error' is 0, indicating success, the operation will be committed
4527 * permanently to the flow table. There is one interesting subcase:
4529 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4530 * the flow table (the "victim" rule) by a new one, then the caller must
4531 * have uninitialized any derived state in the victim rule, as in step 5 in
4532 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4533 * performs steps 6 and 7 for the victim rule, most notably by calling its
4534 * ->rule_dealloc() function.
4536 * If 'error' is nonzero, then generally the operation will be rolled back:
4538 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4539 * restores the original rule. The caller must have uninitialized any
4540 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4541 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4542 * and 7 for the new rule, calling its ->rule_dealloc() function.
4544 * - If 'op' is a "modify flow" operation, ofproto restores the original
4547 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4548 * allowed to fail. It must always succeed.
4550 * Please see the large comment in ofproto/ofproto-provider.h titled
4551 * "Asynchronous Operation Support" for more information. */
4553 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4555 struct ofopgroup *group = op->group;
4557 ovs_assert(op->rule->pending == op);
4558 ovs_assert(group->n_running > 0);
4559 ovs_assert(!error || op->type != OFOPERATION_DELETE);
4562 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4563 ofopgroup_complete(group);
4568 ofoperation_get_victim(struct ofoperation *op)
4570 ovs_assert(op->type == OFOPERATION_ADD);
4575 pick_datapath_id(const struct ofproto *ofproto)
4577 const struct ofport *port;
4579 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4581 uint8_t ea[ETH_ADDR_LEN];
4584 error = netdev_get_etheraddr(port->netdev, ea);
4586 return eth_addr_to_uint64(ea);
4588 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4589 ofproto->name, netdev_get_name(port->netdev),
4592 return ofproto->fallback_dpid;
4596 pick_fallback_dpid(void)
4598 uint8_t ea[ETH_ADDR_LEN];
4599 eth_addr_nicira_random(ea);
4600 return eth_addr_to_uint64(ea);
4603 /* Table overflow policy. */
4605 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4606 * is not configured to evict rules or if the table contains no evictable
4607 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4609 static struct rule *
4610 choose_rule_to_evict(struct oftable *table)
4612 struct eviction_group *evg;
4614 if (!table->eviction_fields) {
4618 /* In the common case, the outer and inner loops here will each be entered
4621 * - The inner loop normally "return"s in its first iteration. If the
4622 * eviction group has any evictable rules, then it always returns in
4625 * - The outer loop only iterates more than once if the largest eviction
4626 * group has no evictable rules.
4628 * - The outer loop can exit only if table's 'max_flows' is all filled up
4629 * by unevictable rules'. */
4630 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4633 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4634 if (rule->evictable) {
4643 /* Searches 'ofproto' for tables that have more flows than their configured
4644 * maximum and that have flow eviction enabled, and evicts as many flows as
4645 * necessary and currently feasible from them.
4647 * This triggers only when an OpenFlow table has N flows in it and then the
4648 * client configures a maximum number of flows less than N. */
4650 ofproto_evict(struct ofproto *ofproto)
4652 struct ofopgroup *group;
4653 struct oftable *table;
4655 group = ofopgroup_create_unattached(ofproto);
4656 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4657 while (classifier_count(&table->cls) > table->max_flows
4658 && table->eviction_fields) {
4661 rule = choose_rule_to_evict(table);
4662 if (!rule || rule->pending) {
4666 ofoperation_create(group, rule,
4667 OFOPERATION_DELETE, OFPRR_EVICTION);
4668 oftable_remove_rule(rule);
4669 ofproto->ofproto_class->rule_destruct(rule);
4672 ofopgroup_submit(group);
4675 /* Eviction groups. */
4677 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4678 * rules. The priority contains low-order random bits to ensure that eviction
4679 * groups with the same number of rules are prioritized randomly. */
4681 eviction_group_priority(size_t n_rules)
4683 uint16_t size = MIN(UINT16_MAX, n_rules);
4684 return (size << 16) | random_uint16();
4687 /* Updates 'evg', an eviction_group within 'table', following a change that
4688 * adds or removes rules in 'evg'. */
4690 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4692 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4693 eviction_group_priority(heap_count(&evg->rules)));
4696 /* Destroys 'evg', an eviction_group within 'table':
4698 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4699 * rules themselves, just removes them from the eviction group.)
4701 * - Removes 'evg' from 'table'.
4705 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4707 while (!heap_is_empty(&evg->rules)) {
4710 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4711 rule->eviction_group = NULL;
4713 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4714 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4715 heap_destroy(&evg->rules);
4719 /* Removes 'rule' from its eviction group, if any. */
4721 eviction_group_remove_rule(struct rule *rule)
4723 if (rule->eviction_group) {
4724 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4725 struct eviction_group *evg = rule->eviction_group;
4727 rule->eviction_group = NULL;
4728 heap_remove(&evg->rules, &rule->evg_node);
4729 if (heap_is_empty(&evg->rules)) {
4730 eviction_group_destroy(table, evg);
4732 eviction_group_resized(table, evg);
4737 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4738 * returns the hash value. */
4740 eviction_group_hash_rule(struct rule *rule)
4742 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4743 const struct mf_subfield *sf;
4747 hash = table->eviction_group_id_basis;
4748 miniflow_expand(&rule->cr.match.flow, &flow);
4749 for (sf = table->eviction_fields;
4750 sf < &table->eviction_fields[table->n_eviction_fields];
4753 if (mf_are_prereqs_ok(sf->field, &flow)) {
4754 union mf_value value;
4756 mf_get_value(sf->field, &flow, &value);
4758 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4760 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4761 unsigned int start = sf->ofs + sf->n_bits;
4762 bitwise_zero(&value, sf->field->n_bytes, start,
4763 sf->field->n_bytes * 8 - start);
4765 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4767 hash = hash_int(hash, 0);
4774 /* Returns an eviction group within 'table' with the given 'id', creating one
4776 static struct eviction_group *
4777 eviction_group_find(struct oftable *table, uint32_t id)
4779 struct eviction_group *evg;
4781 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4785 evg = xmalloc(sizeof *evg);
4786 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4787 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4788 eviction_group_priority(0));
4789 heap_init(&evg->rules);
4794 /* Returns an eviction priority for 'rule'. The return value should be
4795 * interpreted so that higher priorities make a rule more attractive candidates
4798 rule_eviction_priority(struct rule *rule)
4800 long long int hard_expiration;
4801 long long int idle_expiration;
4802 long long int expiration;
4803 uint32_t expiration_offset;
4805 /* Calculate time of expiration. */
4806 hard_expiration = (rule->hard_timeout
4807 ? rule->modified + rule->hard_timeout * 1000
4809 idle_expiration = (rule->idle_timeout
4810 ? rule->used + rule->idle_timeout * 1000
4812 expiration = MIN(hard_expiration, idle_expiration);
4813 if (expiration == LLONG_MAX) {
4817 /* Calculate the time of expiration as a number of (approximate) seconds
4818 * after program startup.
4820 * This should work OK for program runs that last UINT32_MAX seconds or
4821 * less. Therefore, please restart OVS at least once every 136 years. */
4822 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4824 /* Invert the expiration offset because we're using a max-heap. */
4825 return UINT32_MAX - expiration_offset;
4828 /* Adds 'rule' to an appropriate eviction group for its oftable's
4829 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4830 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4833 * The caller must ensure that 'rule' is not already in an eviction group. */
4835 eviction_group_add_rule(struct rule *rule)
4837 struct ofproto *ofproto = rule->ofproto;
4838 struct oftable *table = &ofproto->tables[rule->table_id];
4840 if (table->eviction_fields
4841 && (rule->hard_timeout || rule->idle_timeout)) {
4842 struct eviction_group *evg;
4844 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4846 rule->eviction_group = evg;
4847 heap_insert(&evg->rules, &rule->evg_node,
4848 rule_eviction_priority(rule));
4849 eviction_group_resized(table, evg);
4855 /* Initializes 'table'. */
4857 oftable_init(struct oftable *table)
4859 memset(table, 0, sizeof *table);
4860 classifier_init(&table->cls);
4861 table->max_flows = UINT_MAX;
4864 /* Destroys 'table', including its classifier and eviction groups.
4866 * The caller is responsible for freeing 'table' itself. */
4868 oftable_destroy(struct oftable *table)
4870 ovs_assert(classifier_is_empty(&table->cls));
4871 oftable_disable_eviction(table);
4872 classifier_destroy(&table->cls);
4876 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4877 * string, then 'table' will use its default name.
4879 * This only affects the name exposed for a table exposed through the OpenFlow
4880 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4882 oftable_set_name(struct oftable *table, const char *name)
4884 if (name && name[0]) {
4885 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4886 if (!table->name || strncmp(name, table->name, len)) {
4888 table->name = xmemdup0(name, len);
4896 /* oftables support a choice of two policies when adding a rule would cause the
4897 * number of flows in the table to exceed the configured maximum number: either
4898 * they can refuse to add the new flow or they can evict some existing flow.
4899 * This function configures the former policy on 'table'. */
4901 oftable_disable_eviction(struct oftable *table)
4903 if (table->eviction_fields) {
4904 struct eviction_group *evg, *next;
4906 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4907 &table->eviction_groups_by_id) {
4908 eviction_group_destroy(table, evg);
4910 hmap_destroy(&table->eviction_groups_by_id);
4911 heap_destroy(&table->eviction_groups_by_size);
4913 free(table->eviction_fields);
4914 table->eviction_fields = NULL;
4915 table->n_eviction_fields = 0;
4919 /* oftables support a choice of two policies when adding a rule would cause the
4920 * number of flows in the table to exceed the configured maximum number: either
4921 * they can refuse to add the new flow or they can evict some existing flow.
4922 * This function configures the latter policy on 'table', with fairness based
4923 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4924 * 'n_fields' as 0 disables fairness.) */
4926 oftable_enable_eviction(struct oftable *table,
4927 const struct mf_subfield *fields, size_t n_fields)
4929 struct cls_cursor cursor;
4932 if (table->eviction_fields
4933 && n_fields == table->n_eviction_fields
4935 || !memcmp(fields, table->eviction_fields,
4936 n_fields * sizeof *fields))) {
4941 oftable_disable_eviction(table);
4943 table->n_eviction_fields = n_fields;
4944 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4946 table->eviction_group_id_basis = random_uint32();
4947 hmap_init(&table->eviction_groups_by_id);
4948 heap_init(&table->eviction_groups_by_size);
4950 cls_cursor_init(&cursor, &table->cls, NULL);
4951 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4952 eviction_group_add_rule(rule);
4956 /* Removes 'rule' from the oftable that contains it. */
4958 oftable_remove_rule(struct rule *rule)
4960 struct ofproto *ofproto = rule->ofproto;
4961 struct oftable *table = &ofproto->tables[rule->table_id];
4963 classifier_remove(&table->cls, &rule->cr);
4964 cookies_remove(ofproto, rule);
4965 eviction_group_remove_rule(rule);
4966 if (!list_is_empty(&rule->expirable)) {
4967 list_remove(&rule->expirable);
4971 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4972 * oftable that has an identical cls_rule. Returns the rule that was removed,
4973 * if any, and otherwise NULL. */
4974 static struct rule *
4975 oftable_replace_rule(struct rule *rule)
4977 struct ofproto *ofproto = rule->ofproto;
4978 struct oftable *table = &ofproto->tables[rule->table_id];
4979 struct rule *victim;
4980 bool may_expire = rule->hard_timeout || rule->idle_timeout;
4983 list_insert(&ofproto->expirable, &rule->expirable);
4985 cookies_insert(ofproto, rule);
4987 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4989 cookies_remove(ofproto, victim);
4991 if (!list_is_empty(&victim->expirable)) {
4992 list_remove(&victim->expirable);
4994 eviction_group_remove_rule(victim);
4996 eviction_group_add_rule(rule);
5000 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
5002 oftable_substitute_rule(struct rule *old, struct rule *new)
5005 oftable_replace_rule(new);
5007 oftable_remove_rule(old);
5011 /* unixctl commands. */
5014 ofproto_lookup(const char *name)
5016 struct ofproto *ofproto;
5018 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
5020 if (!strcmp(ofproto->name, name)) {
5028 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
5029 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
5031 struct ofproto *ofproto;
5035 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
5036 ds_put_format(&results, "%s\n", ofproto->name);
5038 unixctl_command_reply(conn, ds_cstr(&results));
5039 ds_destroy(&results);
5043 ofproto_unixctl_init(void)
5045 static bool registered;
5051 unixctl_command_register("ofproto/list", "", 0, 0,
5052 ofproto_unixctl_list, NULL);
5055 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
5057 * This is deprecated. It is only for compatibility with broken device drivers
5058 * in old versions of Linux that do not properly support VLANs when VLAN
5059 * devices are not used. When broken device drivers are no longer in
5060 * widespread use, we will delete these interfaces. */
5062 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
5063 * (exactly) by an OpenFlow rule in 'ofproto'. */
5065 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
5067 const struct oftable *oftable;
5069 free(ofproto->vlan_bitmap);
5070 ofproto->vlan_bitmap = bitmap_allocate(4096);
5071 ofproto->vlans_changed = false;
5073 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
5074 const struct cls_table *table;
5076 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
5077 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
5078 const struct cls_rule *rule;
5080 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
5081 uint16_t vid = miniflow_get_vid(&rule->match.flow);
5082 bitmap_set1(vlan_bitmap, vid);
5083 bitmap_set1(ofproto->vlan_bitmap, vid);
5090 /* Returns true if new VLANs have come into use by the flow table since the
5091 * last call to ofproto_get_vlan_usage().
5093 * We don't track when old VLANs stop being used. */
5095 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
5097 return ofproto->vlans_changed;
5100 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
5101 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
5102 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
5103 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
5104 * then the VLAN device is un-enslaved. */
5106 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
5107 ofp_port_t realdev_ofp_port, int vid)
5109 struct ofport *ofport;
5112 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
5114 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
5116 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
5117 ofproto->name, vlandev_ofp_port);
5121 if (!ofproto->ofproto_class->set_realdev) {
5122 if (!vlandev_ofp_port) {
5125 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
5129 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
5131 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
5132 ofproto->name, vlandev_ofp_port,
5133 netdev_get_name(ofport->netdev), strerror(error));