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.
26 #include "byte-order.h"
27 #include "classifier.h"
30 #include "dynamic-string.h"
33 #include "meta-flow.h"
36 #include "ofp-actions.h"
37 #include "ofp-errors.h"
39 #include "ofp-print.h"
42 #include "ofproto-provider.h"
43 #include "openflow/nicira-ext.h"
44 #include "openflow/openflow.h"
48 #include "poll-loop.h"
54 #include "unaligned.h"
58 VLOG_DEFINE_THIS_MODULE(ofproto);
60 COVERAGE_DEFINE(ofproto_error);
61 COVERAGE_DEFINE(ofproto_flush);
62 COVERAGE_DEFINE(ofproto_no_packet_in);
63 COVERAGE_DEFINE(ofproto_packet_out);
64 COVERAGE_DEFINE(ofproto_queue_req);
65 COVERAGE_DEFINE(ofproto_recv_openflow);
66 COVERAGE_DEFINE(ofproto_reinit_ports);
67 COVERAGE_DEFINE(ofproto_uninstallable);
68 COVERAGE_DEFINE(ofproto_update_port);
71 S_OPENFLOW, /* Processing OpenFlow commands. */
72 S_EVICT, /* Evicting flows from over-limit tables. */
73 S_FLUSH, /* Deleting all flow table rules. */
76 enum ofoperation_type {
83 /* A single OpenFlow request can execute any number of operations. The
84 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
85 * ofconn to which an error reply should be sent if necessary.
87 * ofproto initiates some operations internally. These operations are still
88 * assigned to groups but will not have an associated ofconn. */
90 struct ofproto *ofproto; /* Owning ofproto. */
91 struct list ofproto_node; /* In ofproto's "pending" list. */
92 struct list ops; /* List of "struct ofoperation"s. */
93 int n_running; /* Number of ops still pending. */
95 /* Data needed to send OpenFlow reply on failure or to send a buffered
98 * If list_is_empty(ofconn_node) then this ofopgroup never had an
99 * associated ofconn or its ofconn's connection dropped after it initiated
100 * the operation. In the latter case 'ofconn' is a wild pointer that
101 * refers to freed memory, so the 'ofconn' member must be used only if
102 * !list_is_empty(ofconn_node).
104 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
105 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
106 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
107 uint32_t buffer_id; /* Buffer id from original request. */
110 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
111 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
112 const struct ofp_header *,
114 static void ofopgroup_submit(struct ofopgroup *);
115 static void ofopgroup_complete(struct ofopgroup *);
117 /* A single flow table operation. */
119 struct ofopgroup *group; /* Owning group. */
120 struct list group_node; /* In ofopgroup's "ops" list. */
121 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
122 struct rule *rule; /* Rule being operated upon. */
123 enum ofoperation_type type; /* Type of operation. */
125 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
127 struct rule_actions *actions;
129 /* OFOPERATION_DELETE. */
130 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
132 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
133 uint16_t idle_timeout; /* Rule's old idle timeout. */
134 uint16_t hard_timeout; /* Rule's old hard timeout. */
135 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
136 enum ofperr error; /* 0 if no error. */
139 static struct ofoperation *ofoperation_create(struct ofopgroup *,
141 enum ofoperation_type,
142 enum ofp_flow_removed_reason);
143 static void ofoperation_destroy(struct ofoperation *);
146 static void oftable_init(struct oftable *);
147 static void oftable_destroy(struct oftable *);
149 static void oftable_set_name(struct oftable *, const char *name);
151 static void oftable_disable_eviction(struct oftable *);
152 static void oftable_enable_eviction(struct oftable *,
153 const struct mf_subfield *fields,
156 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
157 static void oftable_remove_rule__(struct ofproto *, struct rule *)
158 OVS_REQUIRES(ofproto_mutex);
159 static void oftable_insert_rule(struct rule *);
161 /* A set of rules within a single OpenFlow table (oftable) that have the same
162 * values for the oftable's eviction_fields. A rule to be evicted, when one is
163 * needed, is taken from the eviction group that contains the greatest number
166 * An oftable owns any number of eviction groups, each of which contains any
169 * Membership in an eviction group is imprecise, based on the hash of the
170 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
171 * That is, if two rules have different eviction_fields, but those
172 * eviction_fields hash to the same value, then they will belong to the same
173 * eviction_group anyway.
175 * (When eviction is not enabled on an oftable, we don't track any eviction
176 * groups, to save time and space.) */
177 struct eviction_group {
178 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
179 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
180 struct heap rules; /* Contains "struct rule"s. */
183 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
184 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static uint32_t rule_eviction_priority(struct rule *);
186 static void eviction_group_add_rule(struct rule *);
187 static void eviction_group_remove_rule(struct rule *);
189 /* Criteria that flow_mod and other operations use for selecting rules on
190 * which to operate. */
191 struct rule_criteria {
192 /* An OpenFlow table or 255 for all tables. */
195 /* OpenFlow matching criteria. Interpreted different in "loose" way by
196 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
197 * defined in the OpenFlow spec. */
200 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
201 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
202 * The rule will not be selected if M is 1 and B != C. */
204 ovs_be64 cookie_mask;
206 /* Selection based on actions within a rule:
208 * If out_port != OFPP_ANY, selects only rules that output to out_port.
209 * If out_group != OFPG_ALL, select only rules that output to out_group. */
214 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
215 const struct match *match,
216 unsigned int priority,
217 ovs_be64 cookie, ovs_be64 cookie_mask,
218 ofp_port_t out_port, uint32_t out_group);
219 static void rule_criteria_destroy(struct rule_criteria *);
221 /* A packet that needs to be passed to rule_execute().
223 * (We can't do this immediately from ofopgroup_complete() because that holds
224 * ofproto_mutex, which rule_execute() needs released.) */
225 struct rule_execute {
226 struct list list_node; /* In struct ofproto's "rule_executes" list. */
227 struct rule *rule; /* Owns a reference to the rule. */
229 struct ofpbuf *packet; /* Owns the packet. */
232 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
233 static void destroy_rule_executes(struct ofproto *);
236 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
237 static void ofport_destroy(struct ofport *);
239 static void update_port(struct ofproto *, const char *devname);
240 static int init_ports(struct ofproto *);
241 static void reinit_ports(struct ofproto *);
244 static void ofproto_rule_destroy__(struct rule *);
245 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
246 static bool rule_is_modifiable(const struct rule *);
249 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
250 struct ofputil_flow_mod *,
251 const struct ofp_header *);
252 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
253 struct ofputil_flow_mod *,
254 const struct ofp_header *,
255 const struct rule_collection *);
256 static void delete_flow__(struct rule *rule, struct ofopgroup *,
257 enum ofp_flow_removed_reason)
258 OVS_REQUIRES(ofproto_mutex);
259 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
260 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
261 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
262 struct ofputil_flow_mod *,
263 const struct ofp_header *)
264 OVS_EXCLUDED(ofproto_mutex);
265 static void calc_duration(long long int start, long long int now,
266 uint32_t *sec, uint32_t *nsec);
269 static uint64_t pick_datapath_id(const struct ofproto *);
270 static uint64_t pick_fallback_dpid(void);
271 static void ofproto_destroy__(struct ofproto *);
272 static void update_mtu(struct ofproto *, struct ofport *);
273 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
276 static void ofproto_unixctl_init(void);
278 /* All registered ofproto classes, in probe order. */
279 static const struct ofproto_class **ofproto_classes;
280 static size_t n_ofproto_classes;
281 static size_t allocated_ofproto_classes;
283 /* Global lock that protects all flow table operations. */
284 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
286 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
287 unsigned n_handler_threads;
288 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
290 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
291 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
293 /* Initial mappings of port to OpenFlow number mappings. */
294 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
296 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
298 /* The default value of true waits for flow restore. */
299 static bool flow_restore_wait = true;
301 /* Must be called to initialize the ofproto library.
303 * The caller may pass in 'iface_hints', which contains an shash of
304 * "iface_hint" elements indexed by the interface's name. The provider
305 * may use these hints to describe the startup configuration in order to
306 * reinitialize its state. The caller owns the provided data, so a
307 * provider will make copies of anything required. An ofproto provider
308 * will remove any existing state that is not described by the hint, and
309 * may choose to remove it all. */
311 ofproto_init(const struct shash *iface_hints)
313 struct shash_node *node;
316 ofproto_class_register(&ofproto_dpif_class);
318 /* Make a local copy, since we don't own 'iface_hints' elements. */
319 SHASH_FOR_EACH(node, iface_hints) {
320 const struct iface_hint *orig_hint = node->data;
321 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
322 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
324 new_hint->br_name = xstrdup(orig_hint->br_name);
325 new_hint->br_type = xstrdup(br_type);
326 new_hint->ofp_port = orig_hint->ofp_port;
328 shash_add(&init_ofp_ports, node->name, new_hint);
331 for (i = 0; i < n_ofproto_classes; i++) {
332 ofproto_classes[i]->init(&init_ofp_ports);
336 /* 'type' should be a normalized datapath type, as returned by
337 * ofproto_normalize_type(). Returns the corresponding ofproto_class
338 * structure, or a null pointer if there is none registered for 'type'. */
339 static const struct ofproto_class *
340 ofproto_class_find__(const char *type)
344 for (i = 0; i < n_ofproto_classes; i++) {
345 const struct ofproto_class *class = ofproto_classes[i];
350 class->enumerate_types(&types);
351 found = sset_contains(&types, type);
352 sset_destroy(&types);
358 VLOG_WARN("unknown datapath type %s", type);
362 /* Registers a new ofproto class. After successful registration, new ofprotos
363 * of that type can be created using ofproto_create(). */
365 ofproto_class_register(const struct ofproto_class *new_class)
369 for (i = 0; i < n_ofproto_classes; i++) {
370 if (ofproto_classes[i] == new_class) {
375 if (n_ofproto_classes >= allocated_ofproto_classes) {
376 ofproto_classes = x2nrealloc(ofproto_classes,
377 &allocated_ofproto_classes,
378 sizeof *ofproto_classes);
380 ofproto_classes[n_ofproto_classes++] = new_class;
384 /* Unregisters a datapath provider. 'type' must have been previously
385 * registered and not currently be in use by any ofprotos. After
386 * unregistration new datapaths of that type cannot be opened using
387 * ofproto_create(). */
389 ofproto_class_unregister(const struct ofproto_class *class)
393 for (i = 0; i < n_ofproto_classes; i++) {
394 if (ofproto_classes[i] == class) {
395 for (i++; i < n_ofproto_classes; i++) {
396 ofproto_classes[i - 1] = ofproto_classes[i];
402 VLOG_WARN("attempted to unregister an ofproto class that is not "
407 /* Clears 'types' and enumerates all registered ofproto types into it. The
408 * caller must first initialize the sset. */
410 ofproto_enumerate_types(struct sset *types)
414 for (i = 0; i < n_ofproto_classes; i++) {
415 ofproto_classes[i]->enumerate_types(types);
419 /* Returns the fully spelled out name for the given ofproto 'type'.
421 * Normalized type string can be compared with strcmp(). Unnormalized type
422 * string might be the same even if they have different spellings. */
424 ofproto_normalize_type(const char *type)
426 return type && type[0] ? type : "system";
429 /* Clears 'names' and enumerates the names of all known created ofprotos with
430 * the given 'type'. The caller must first initialize the sset. Returns 0 if
431 * successful, otherwise a positive errno value.
433 * Some kinds of datapaths might not be practically enumerable. This is not
434 * considered an error. */
436 ofproto_enumerate_names(const char *type, struct sset *names)
438 const struct ofproto_class *class = ofproto_class_find__(type);
439 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
443 ofproto_create(const char *datapath_name, const char *datapath_type,
444 struct ofproto **ofprotop)
446 const struct ofproto_class *class;
447 struct ofproto *ofproto;
453 ofproto_unixctl_init();
455 datapath_type = ofproto_normalize_type(datapath_type);
456 class = ofproto_class_find__(datapath_type);
458 VLOG_WARN("could not create datapath %s of unknown type %s",
459 datapath_name, datapath_type);
463 ofproto = class->alloc();
465 VLOG_ERR("failed to allocate datapath %s of type %s",
466 datapath_name, datapath_type);
471 ovs_mutex_lock(&ofproto_mutex);
472 memset(ofproto, 0, sizeof *ofproto);
473 ofproto->ofproto_class = class;
474 ofproto->name = xstrdup(datapath_name);
475 ofproto->type = xstrdup(datapath_type);
476 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
477 hash_string(ofproto->name, 0));
478 ofproto->datapath_id = 0;
479 ofproto->forward_bpdu = false;
480 ofproto->fallback_dpid = pick_fallback_dpid();
481 ofproto->mfr_desc = NULL;
482 ofproto->hw_desc = NULL;
483 ofproto->sw_desc = NULL;
484 ofproto->serial_desc = NULL;
485 ofproto->dp_desc = NULL;
486 ofproto->frag_handling = OFPC_FRAG_NORMAL;
487 hmap_init(&ofproto->ports);
488 shash_init(&ofproto->port_by_name);
489 simap_init(&ofproto->ofp_requests);
490 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
491 ofproto->eviction_group_timer = LLONG_MIN;
492 ofproto->tables = NULL;
493 ofproto->n_tables = 0;
494 hindex_init(&ofproto->cookies);
495 list_init(&ofproto->expirable);
496 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
497 ofproto->state = S_OPENFLOW;
498 list_init(&ofproto->pending);
499 ofproto->n_pending = 0;
500 hmap_init(&ofproto->deletions);
501 guarded_list_init(&ofproto->rule_executes);
502 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
503 ofproto->first_op = ofproto->last_op = LLONG_MIN;
504 ofproto->next_op_report = LLONG_MAX;
505 ofproto->op_backoff = LLONG_MIN;
506 ofproto->vlan_bitmap = NULL;
507 ofproto->vlans_changed = false;
508 ofproto->min_mtu = INT_MAX;
509 ovs_rwlock_init(&ofproto->groups_rwlock);
510 hmap_init(&ofproto->groups);
511 ovs_mutex_unlock(&ofproto_mutex);
513 error = ofproto->ofproto_class->construct(ofproto);
515 VLOG_ERR("failed to open datapath %s: %s",
516 datapath_name, ovs_strerror(error));
517 ofproto_destroy__(ofproto);
521 /* The "max_ports" member should have been set by ->construct(ofproto).
522 * Port 0 is not a valid OpenFlow port, so mark that as unavailable. */
523 ofproto->ofp_port_ids = bitmap_allocate(ofproto->max_ports);
524 bitmap_set1(ofproto->ofp_port_ids, 0);
526 /* Check that hidden tables, if any, are at the end. */
527 ovs_assert(ofproto->n_tables);
528 for (i = 0; i + 1 < ofproto->n_tables; i++) {
529 enum oftable_flags flags = ofproto->tables[i].flags;
530 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
532 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
535 ofproto->datapath_id = pick_datapath_id(ofproto);
538 /* Initialize meters table. */
539 if (ofproto->ofproto_class->meter_get_features) {
540 ofproto->ofproto_class->meter_get_features(ofproto,
541 &ofproto->meter_features);
543 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
545 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
546 * sizeof(struct meter *));
552 /* Must be called (only) by an ofproto implementation in its constructor
553 * function. See the large comment on 'construct' in struct ofproto_class for
556 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
558 struct oftable *table;
560 ovs_assert(!ofproto->n_tables);
561 ovs_assert(n_tables >= 1 && n_tables <= 255);
563 ofproto->n_tables = n_tables;
564 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
565 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
570 /* To be optionally called (only) by an ofproto implementation in its
571 * constructor function. See the large comment on 'construct' in struct
572 * ofproto_class for details.
574 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
575 * will then ensure that actions passed into the ofproto implementation will
576 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
577 * function is not called, there will be no such restriction.
579 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
580 * the 'max_ports' restriction. */
582 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
584 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
585 ofproto->max_ports = max_ports;
589 ofproto_get_datapath_id(const struct ofproto *ofproto)
591 return ofproto->datapath_id;
595 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
597 uint64_t old_dpid = p->datapath_id;
598 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
599 if (p->datapath_id != old_dpid) {
600 /* Force all active connections to reconnect, since there is no way to
601 * notify a controller that the datapath ID has changed. */
602 ofproto_reconnect_controllers(p);
607 ofproto_set_controllers(struct ofproto *p,
608 const struct ofproto_controller *controllers,
609 size_t n_controllers, uint32_t allowed_versions)
611 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
616 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
618 connmgr_set_fail_mode(p->connmgr, fail_mode);
621 /* Drops the connections between 'ofproto' and all of its controllers, forcing
622 * them to reconnect. */
624 ofproto_reconnect_controllers(struct ofproto *ofproto)
626 connmgr_reconnect(ofproto->connmgr);
629 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
630 * in-band control should guarantee access, in the same way that in-band
631 * control guarantees access to OpenFlow controllers. */
633 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
634 const struct sockaddr_in *extras, size_t n)
636 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
639 /* Sets the OpenFlow queue used by flows set up by in-band control on
640 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
641 * flows will use the default queue. */
643 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
645 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
648 /* Sets the number of flows at which eviction from the kernel flow table
651 ofproto_set_flow_eviction_threshold(unsigned threshold)
653 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
657 /* Sets the path for handling flow misses. */
659 ofproto_set_flow_miss_model(unsigned model)
661 flow_miss_model = model;
664 /* If forward_bpdu is true, the NORMAL action will forward frames with
665 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
666 * the NORMAL action will drop these frames. */
668 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
670 bool old_val = ofproto->forward_bpdu;
671 ofproto->forward_bpdu = forward_bpdu;
672 if (old_val != ofproto->forward_bpdu) {
673 if (ofproto->ofproto_class->forward_bpdu_changed) {
674 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
679 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
680 * 'idle_time', in seconds, and the maximum number of MAC table entries to
683 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
686 if (ofproto->ofproto_class->set_mac_table_config) {
687 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
692 /* Sets number of upcall handler threads. The default is
693 * (number of online cores - 2). */
695 ofproto_set_n_handler_threads(unsigned limit)
698 n_handler_threads = limit;
700 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
701 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
706 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
709 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
713 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
715 return connmgr_set_snoops(ofproto->connmgr, snoops);
719 ofproto_set_netflow(struct ofproto *ofproto,
720 const struct netflow_options *nf_options)
722 if (nf_options && sset_is_empty(&nf_options->collectors)) {
726 if (ofproto->ofproto_class->set_netflow) {
727 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
729 return nf_options ? EOPNOTSUPP : 0;
734 ofproto_set_sflow(struct ofproto *ofproto,
735 const struct ofproto_sflow_options *oso)
737 if (oso && sset_is_empty(&oso->targets)) {
741 if (ofproto->ofproto_class->set_sflow) {
742 return ofproto->ofproto_class->set_sflow(ofproto, oso);
744 return oso ? EOPNOTSUPP : 0;
749 ofproto_set_ipfix(struct ofproto *ofproto,
750 const struct ofproto_ipfix_bridge_exporter_options *bo,
751 const struct ofproto_ipfix_flow_exporter_options *fo,
754 if (ofproto->ofproto_class->set_ipfix) {
755 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
757 return (bo || fo) ? EOPNOTSUPP : 0;
762 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
764 flow_restore_wait = flow_restore_wait_db;
768 ofproto_get_flow_restore_wait(void)
770 return flow_restore_wait;
774 /* Spanning Tree Protocol (STP) configuration. */
776 /* Configures STP on 'ofproto' using the settings defined in 's'. If
777 * 's' is NULL, disables STP.
779 * Returns 0 if successful, otherwise a positive errno value. */
781 ofproto_set_stp(struct ofproto *ofproto,
782 const struct ofproto_stp_settings *s)
784 return (ofproto->ofproto_class->set_stp
785 ? ofproto->ofproto_class->set_stp(ofproto, s)
789 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
790 * 'enabled' member of 's' is false, then the other members are not
793 * Returns 0 if successful, otherwise a positive errno value. */
795 ofproto_get_stp_status(struct ofproto *ofproto,
796 struct ofproto_stp_status *s)
798 return (ofproto->ofproto_class->get_stp_status
799 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
803 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
804 * in 's'. The caller is responsible for assigning STP port numbers
805 * (using the 'port_num' member in the range of 1 through 255, inclusive)
806 * and ensuring there are no duplicates. If the 's' is NULL, then STP
807 * is disabled on the port.
809 * Returns 0 if successful, otherwise a positive errno value.*/
811 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
812 const struct ofproto_port_stp_settings *s)
814 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
816 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
817 ofproto->name, ofp_port);
821 return (ofproto->ofproto_class->set_stp_port
822 ? ofproto->ofproto_class->set_stp_port(ofport, s)
826 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
827 * 's'. If the 'enabled' member in 's' is false, then the other members
828 * are not meaningful.
830 * Returns 0 if successful, otherwise a positive errno value.*/
832 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
833 struct ofproto_port_stp_status *s)
835 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
837 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
838 "port %"PRIu16, ofproto->name, ofp_port);
842 return (ofproto->ofproto_class->get_stp_port_status
843 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
847 /* Queue DSCP configuration. */
849 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
850 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
851 * to implement QoS. Instead, it is used to implement features which require
852 * knowledge of what queues exist on a port, and some basic information about
855 * Returns 0 if successful, otherwise a positive errno value. */
857 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
858 const struct ofproto_port_queue *queues,
861 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
864 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
865 ofproto->name, ofp_port);
869 return (ofproto->ofproto_class->set_queues
870 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
874 /* Connectivity Fault Management configuration. */
876 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
878 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
880 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
881 if (ofport && ofproto->ofproto_class->set_cfm) {
882 ofproto->ofproto_class->set_cfm(ofport, NULL);
886 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
887 * basic configuration from the configuration members in 'cfm', and the remote
888 * maintenance point ID from remote_mpid. Ignores the statistics members of
891 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
893 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
894 const struct cfm_settings *s)
896 struct ofport *ofport;
899 ofport = ofproto_get_port(ofproto, ofp_port);
901 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
902 ofproto->name, ofp_port);
906 /* XXX: For configuration simplicity, we only support one remote_mpid
907 * outside of the CFM module. It's not clear if this is the correct long
908 * term solution or not. */
909 error = (ofproto->ofproto_class->set_cfm
910 ? ofproto->ofproto_class->set_cfm(ofport, s)
913 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
914 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
915 ovs_strerror(error));
919 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
920 * 'ofproto' does not have a port 'ofp_port'. */
922 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
923 const struct smap *cfg)
925 struct ofport *ofport;
928 ofport = ofproto_get_port(ofproto, ofp_port);
930 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
931 ofproto->name, ofp_port);
935 error = (ofproto->ofproto_class->set_bfd
936 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
939 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
940 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
941 ovs_strerror(error));
945 /* Populates 'status' with key value pairs indicating the status of the BFD
946 * session on 'ofp_port'. This information is intended to be populated in the
947 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
950 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
953 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
954 return (ofport && ofproto->ofproto_class->get_bfd_status
955 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
959 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
960 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
961 * 0 if LACP partner information is not current (generally indicating a
962 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
964 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
966 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
967 return (ofport && ofproto->ofproto_class->port_is_lacp_current
968 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
974 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
975 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
976 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
977 * configuration plus, if there is more than one slave, a bonding
980 * If 'aux' is already registered then this function updates its configuration
981 * to 's'. Otherwise, this function registers a new bundle.
983 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
986 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
987 const struct ofproto_bundle_settings *s)
989 return (ofproto->ofproto_class->bundle_set
990 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
994 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
995 * If no such bundle has been registered, this has no effect. */
997 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
999 return ofproto_bundle_register(ofproto, aux, NULL);
1003 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1004 * If 'aux' is already registered then this function updates its configuration
1005 * to 's'. Otherwise, this function registers a new mirror. */
1007 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1008 const struct ofproto_mirror_settings *s)
1010 return (ofproto->ofproto_class->mirror_set
1011 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1015 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1016 * If no mirror has been registered, this has no effect. */
1018 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1020 return ofproto_mirror_register(ofproto, aux, NULL);
1023 /* Retrieves statistics from mirror associated with client data pointer
1024 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1025 * 'bytes', respectively. If a particular counters is not supported,
1026 * the appropriate argument is set to UINT64_MAX. */
1028 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1029 uint64_t *packets, uint64_t *bytes)
1031 if (!ofproto->ofproto_class->mirror_get_stats) {
1032 *packets = *bytes = UINT64_MAX;
1036 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1040 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1041 * which all packets are flooded, instead of using MAC learning. If
1042 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1044 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1047 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1049 return (ofproto->ofproto_class->set_flood_vlans
1050 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1054 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1055 * output for a mirror. */
1057 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1059 return (ofproto->ofproto_class->is_mirror_output_bundle
1060 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1064 /* Configuration of OpenFlow tables. */
1066 /* Returns the number of OpenFlow tables in 'ofproto'. */
1068 ofproto_get_n_tables(const struct ofproto *ofproto)
1070 return ofproto->n_tables;
1073 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1074 * settings from 's'. 'table_id' must be in the range 0 through the number of
1075 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1077 * For read-only tables, only the name may be configured. */
1079 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1080 const struct ofproto_table_settings *s)
1082 struct oftable *table;
1084 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1085 table = &ofproto->tables[table_id];
1087 oftable_set_name(table, s->name);
1089 if (table->flags & OFTABLE_READONLY) {
1094 oftable_enable_eviction(table, s->groups, s->n_groups);
1096 oftable_disable_eviction(table);
1099 table->max_flows = s->max_flows;
1100 ovs_rwlock_rdlock(&table->cls.rwlock);
1101 if (classifier_count(&table->cls) > table->max_flows
1102 && table->eviction_fields) {
1103 /* 'table' contains more flows than allowed. We might not be able to
1104 * evict them right away because of the asynchronous nature of flow
1105 * table changes. Schedule eviction for later. */
1106 switch (ofproto->state) {
1108 ofproto->state = S_EVICT;
1112 /* We're already deleting flows, nothing more to do. */
1116 ovs_rwlock_unlock(&table->cls.rwlock);
1120 ofproto_has_snoops(const struct ofproto *ofproto)
1122 return connmgr_has_snoops(ofproto->connmgr);
1126 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1128 connmgr_get_snoops(ofproto->connmgr, snoops);
1132 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule)
1133 OVS_REQUIRES(ofproto_mutex)
1135 struct ofopgroup *group;
1137 ovs_assert(!rule->pending);
1139 group = ofopgroup_create_unattached(ofproto);
1140 delete_flow__(rule, group, OFPRR_DELETE);
1141 ofopgroup_submit(group);
1144 /* Deletes 'rule' from 'cls' within 'ofproto'.
1146 * Within an ofproto implementation, this function allows an ofproto
1147 * implementation to destroy any rules that remain when its ->destruct()
1148 * function is called. This function is not suitable for use elsewhere in an
1149 * ofproto implementation.
1151 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1152 * Cycle" in ofproto-provider.h. */
1154 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1155 OVS_EXCLUDED(ofproto_mutex)
1157 struct ofopgroup *group;
1159 ovs_mutex_lock(&ofproto_mutex);
1160 ovs_assert(!rule->pending);
1162 group = ofopgroup_create_unattached(ofproto);
1163 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1164 oftable_remove_rule__(ofproto, rule);
1165 ofproto->ofproto_class->rule_delete(rule);
1166 ofopgroup_submit(group);
1168 ovs_mutex_unlock(&ofproto_mutex);
1172 ofproto_flush__(struct ofproto *ofproto)
1173 OVS_EXCLUDED(ofproto_mutex)
1175 struct oftable *table;
1177 if (ofproto->ofproto_class->flush) {
1178 ofproto->ofproto_class->flush(ofproto);
1181 ovs_mutex_lock(&ofproto_mutex);
1182 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1183 struct rule *rule, *next_rule;
1184 struct cls_cursor cursor;
1186 if (table->flags & OFTABLE_HIDDEN) {
1190 ovs_rwlock_rdlock(&table->cls.rwlock);
1191 cls_cursor_init(&cursor, &table->cls, NULL);
1192 ovs_rwlock_unlock(&table->cls.rwlock);
1193 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1194 if (!rule->pending) {
1195 ofproto_rule_delete__(ofproto, rule);
1199 ovs_mutex_unlock(&ofproto_mutex);
1202 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1205 ofproto_destroy__(struct ofproto *ofproto)
1206 OVS_EXCLUDED(ofproto_mutex)
1208 struct oftable *table;
1210 ovs_assert(list_is_empty(&ofproto->pending));
1212 destroy_rule_executes(ofproto);
1213 guarded_list_destroy(&ofproto->rule_executes);
1215 delete_group(ofproto, OFPG_ALL);
1216 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1217 hmap_destroy(&ofproto->groups);
1219 connmgr_destroy(ofproto->connmgr);
1221 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1222 free(ofproto->name);
1223 free(ofproto->type);
1224 free(ofproto->mfr_desc);
1225 free(ofproto->hw_desc);
1226 free(ofproto->sw_desc);
1227 free(ofproto->serial_desc);
1228 free(ofproto->dp_desc);
1229 hmap_destroy(&ofproto->ports);
1230 shash_destroy(&ofproto->port_by_name);
1231 bitmap_free(ofproto->ofp_port_ids);
1232 simap_destroy(&ofproto->ofp_requests);
1234 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1235 oftable_destroy(table);
1237 free(ofproto->tables);
1239 hmap_destroy(&ofproto->deletions);
1241 free(ofproto->vlan_bitmap);
1243 ofproto->ofproto_class->dealloc(ofproto);
1247 ofproto_destroy(struct ofproto *p)
1248 OVS_EXCLUDED(ofproto_mutex)
1250 struct ofport *ofport, *next_ofport;
1257 meter_delete(p, 1, p->meter_features.max_meters);
1258 p->meter_features.max_meters = 0;
1264 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1265 ofport_destroy(ofport);
1268 p->ofproto_class->destruct(p);
1269 ofproto_destroy__(p);
1272 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1273 * kernel datapath, for example, this destroys the datapath in the kernel, and
1274 * with the netdev-based datapath, it tears down the data structures that
1275 * represent the datapath.
1277 * The datapath should not be currently open as an ofproto. */
1279 ofproto_delete(const char *name, const char *type)
1281 const struct ofproto_class *class = ofproto_class_find__(type);
1282 return (!class ? EAFNOSUPPORT
1283 : !class->del ? EACCES
1284 : class->del(type, name));
1288 process_port_change(struct ofproto *ofproto, int error, char *devname)
1290 if (error == ENOBUFS) {
1291 reinit_ports(ofproto);
1292 } else if (!error) {
1293 update_port(ofproto, devname);
1299 ofproto_type_run(const char *datapath_type)
1301 const struct ofproto_class *class;
1304 datapath_type = ofproto_normalize_type(datapath_type);
1305 class = ofproto_class_find__(datapath_type);
1307 error = class->type_run ? class->type_run(datapath_type) : 0;
1308 if (error && error != EAGAIN) {
1309 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1310 datapath_type, ovs_strerror(error));
1316 ofproto_type_run_fast(const char *datapath_type)
1318 const struct ofproto_class *class;
1321 datapath_type = ofproto_normalize_type(datapath_type);
1322 class = ofproto_class_find__(datapath_type);
1324 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1325 if (error && error != EAGAIN) {
1326 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1327 datapath_type, ovs_strerror(error));
1333 ofproto_type_wait(const char *datapath_type)
1335 const struct ofproto_class *class;
1337 datapath_type = ofproto_normalize_type(datapath_type);
1338 class = ofproto_class_find__(datapath_type);
1340 if (class->type_wait) {
1341 class->type_wait(datapath_type);
1346 any_pending_ops(const struct ofproto *p)
1347 OVS_EXCLUDED(ofproto_mutex)
1351 ovs_mutex_lock(&ofproto_mutex);
1352 b = !list_is_empty(&p->pending);
1353 ovs_mutex_unlock(&ofproto_mutex);
1359 ofproto_run(struct ofproto *p)
1361 struct sset changed_netdevs;
1362 const char *changed_netdev;
1363 struct ofport *ofport;
1366 error = p->ofproto_class->run(p);
1367 if (error && error != EAGAIN) {
1368 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1371 run_rule_executes(p);
1373 /* Restore the eviction group heap invariant occasionally. */
1374 if (p->eviction_group_timer < time_msec()) {
1377 p->eviction_group_timer = time_msec() + 1000;
1379 for (i = 0; i < p->n_tables; i++) {
1380 struct oftable *table = &p->tables[i];
1381 struct eviction_group *evg;
1382 struct cls_cursor cursor;
1385 if (!table->eviction_fields) {
1389 ovs_mutex_lock(&ofproto_mutex);
1390 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1391 heap_rebuild(&evg->rules);
1394 ovs_rwlock_rdlock(&table->cls.rwlock);
1395 cls_cursor_init(&cursor, &table->cls, NULL);
1396 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1397 if (!rule->eviction_group
1398 && (rule->idle_timeout || rule->hard_timeout)) {
1399 eviction_group_add_rule(rule);
1402 ovs_rwlock_unlock(&table->cls.rwlock);
1403 ovs_mutex_unlock(&ofproto_mutex);
1407 if (p->ofproto_class->port_poll) {
1410 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1411 process_port_change(p, error, devname);
1415 /* Update OpenFlow port status for any port whose netdev has changed.
1417 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1418 * destroyed, so it's not safe to update ports directly from the
1419 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1420 * need this two-phase approach. */
1421 sset_init(&changed_netdevs);
1422 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1423 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1424 if (ofport->change_seq != change_seq) {
1425 ofport->change_seq = change_seq;
1426 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1429 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1430 update_port(p, changed_netdev);
1432 sset_destroy(&changed_netdevs);
1436 connmgr_run(p->connmgr, handle_openflow);
1440 connmgr_run(p->connmgr, NULL);
1442 if (!any_pending_ops(p)) {
1443 p->state = S_OPENFLOW;
1448 connmgr_run(p->connmgr, NULL);
1450 if (!any_pending_ops(p)) {
1451 connmgr_flushed(p->connmgr);
1452 p->state = S_OPENFLOW;
1460 if (time_msec() >= p->next_op_report) {
1461 long long int ago = (time_msec() - p->first_op) / 1000;
1462 long long int interval = (p->last_op - p->first_op) / 1000;
1466 ds_put_format(&s, "%d flow_mods ",
1467 p->n_add + p->n_delete + p->n_modify);
1468 if (interval == ago) {
1469 ds_put_format(&s, "in the last %lld s", ago);
1470 } else if (interval) {
1471 ds_put_format(&s, "in the %lld s starting %lld s ago",
1474 ds_put_format(&s, "%lld s ago", ago);
1477 ds_put_cstr(&s, " (");
1479 ds_put_format(&s, "%d adds, ", p->n_add);
1482 ds_put_format(&s, "%d deletes, ", p->n_delete);
1485 ds_put_format(&s, "%d modifications, ", p->n_modify);
1488 ds_put_char(&s, ')');
1490 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1493 p->n_add = p->n_delete = p->n_modify = 0;
1494 p->next_op_report = LLONG_MAX;
1500 /* Performs periodic activity required by 'ofproto' that needs to be done
1501 * with the least possible latency.
1503 * It makes sense to call this function a couple of times per poll loop, to
1504 * provide a significant performance boost on some benchmarks with the
1505 * ofproto-dpif implementation. */
1507 ofproto_run_fast(struct ofproto *p)
1511 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1512 if (error && error != EAGAIN) {
1513 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1514 p->name, ovs_strerror(error));
1520 ofproto_wait(struct ofproto *p)
1522 struct ofport *ofport;
1524 p->ofproto_class->wait(p);
1525 if (p->ofproto_class->port_poll_wait) {
1526 p->ofproto_class->port_poll_wait(p);
1529 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1530 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1531 poll_immediate_wake();
1537 connmgr_wait(p->connmgr, true);
1542 connmgr_wait(p->connmgr, false);
1543 if (!any_pending_ops(p)) {
1544 poll_immediate_wake();
1551 ofproto_is_alive(const struct ofproto *p)
1553 return connmgr_has_controllers(p->connmgr);
1556 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1557 * memory_report(). */
1559 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1561 const struct oftable *table;
1562 unsigned int n_rules;
1564 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1566 ovs_mutex_lock(&ofproto_mutex);
1567 simap_increase(usage, "ops",
1568 ofproto->n_pending + hmap_count(&ofproto->deletions));
1569 ovs_mutex_unlock(&ofproto_mutex);
1572 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1573 ovs_rwlock_rdlock(&table->cls.rwlock);
1574 n_rules += classifier_count(&table->cls);
1575 ovs_rwlock_unlock(&table->cls.rwlock);
1577 simap_increase(usage, "rules", n_rules);
1579 if (ofproto->ofproto_class->get_memory_usage) {
1580 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1583 connmgr_get_memory_usage(ofproto->connmgr, usage);
1587 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1590 connmgr_get_controller_info(ofproto->connmgr, info);
1594 ofproto_free_ofproto_controller_info(struct shash *info)
1596 connmgr_free_controller_info(info);
1599 /* Makes a deep copy of 'old' into 'port'. */
1601 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1603 port->name = xstrdup(old->name);
1604 port->type = xstrdup(old->type);
1605 port->ofp_port = old->ofp_port;
1608 /* Frees memory allocated to members of 'ofproto_port'.
1610 * Do not call this function on an ofproto_port obtained from
1611 * ofproto_port_dump_next(): that function retains ownership of the data in the
1614 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1616 free(ofproto_port->name);
1617 free(ofproto_port->type);
1620 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1622 * This function provides no status indication. An error status for the entire
1623 * dump operation is provided when it is completed by calling
1624 * ofproto_port_dump_done().
1627 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1628 const struct ofproto *ofproto)
1630 dump->ofproto = ofproto;
1631 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1635 /* Attempts to retrieve another port from 'dump', which must have been created
1636 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1637 * 'port' and returns true. On failure, returns false.
1639 * Failure might indicate an actual error or merely that the last port has been
1640 * dumped. An error status for the entire dump operation is provided when it
1641 * is completed by calling ofproto_port_dump_done().
1643 * The ofproto owns the data stored in 'port'. It will remain valid until at
1644 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1645 * ofproto_port_dump_done(). */
1647 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1648 struct ofproto_port *port)
1650 const struct ofproto *ofproto = dump->ofproto;
1656 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1659 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1665 /* Completes port table dump operation 'dump', which must have been created
1666 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1667 * error-free, otherwise a positive errno value describing the problem. */
1669 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1671 const struct ofproto *ofproto = dump->ofproto;
1673 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1676 return dump->error == EOF ? 0 : dump->error;
1679 /* Returns the type to pass to netdev_open() when a datapath of type
1680 * 'datapath_type' has a port of type 'port_type', for a few special
1681 * cases when a netdev type differs from a port type. For example, when
1682 * using the userspace datapath, a port of type "internal" needs to be
1685 * Returns either 'type' itself or a string literal, which must not be
1688 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1690 const struct ofproto_class *class;
1692 datapath_type = ofproto_normalize_type(datapath_type);
1693 class = ofproto_class_find__(datapath_type);
1698 return (class->port_open_type
1699 ? class->port_open_type(datapath_type, port_type)
1703 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1704 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1705 * the port's OpenFlow port number.
1707 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1708 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1709 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1710 * 'ofp_portp' is non-null). */
1712 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1713 ofp_port_t *ofp_portp)
1715 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1718 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1720 const char *netdev_name = netdev_get_name(netdev);
1722 simap_put(&ofproto->ofp_requests, netdev_name,
1723 ofp_to_u16(ofp_port));
1724 update_port(ofproto, netdev_name);
1727 struct ofproto_port ofproto_port;
1729 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1731 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1732 ofproto_port_destroy(&ofproto_port);
1737 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1738 * initializes '*port' appropriately; on failure, returns a positive errno
1741 * The caller owns the data in 'ofproto_port' and must free it with
1742 * ofproto_port_destroy() when it is no longer needed. */
1744 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1745 struct ofproto_port *port)
1749 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1751 memset(port, 0, sizeof *port);
1756 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1757 * Returns 0 if successful, otherwise a positive errno. */
1759 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1761 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1762 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1763 struct simap_node *ofp_request_node;
1766 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1767 if (ofp_request_node) {
1768 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1771 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1772 if (!error && ofport) {
1773 /* 'name' is the netdev's name and update_port() is going to close the
1774 * netdev. Just in case update_port() refers to 'name' after it
1775 * destroys 'ofport', make a copy of it around the update_port()
1777 char *devname = xstrdup(name);
1778 update_port(ofproto, devname);
1785 simple_flow_mod(struct ofproto *ofproto,
1786 const struct match *match, unsigned int priority,
1787 const struct ofpact *ofpacts, size_t ofpacts_len,
1788 enum ofp_flow_mod_command command)
1790 struct ofputil_flow_mod fm;
1792 memset(&fm, 0, sizeof fm);
1794 fm.priority = priority;
1797 fm.modify_cookie = false;
1799 fm.command = command;
1800 fm.idle_timeout = 0;
1801 fm.hard_timeout = 0;
1802 fm.buffer_id = UINT32_MAX;
1803 fm.out_port = OFPP_ANY;
1804 fm.out_group = OFPG_ANY;
1806 fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1807 fm.ofpacts_len = ofpacts_len;
1809 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1812 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1813 * performs the 'n_actions' actions in 'actions'. The new flow will not
1816 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1817 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1818 * controllers; otherwise, it will be hidden.
1820 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1822 * This is a helper function for in-band control and fail-open. */
1824 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1825 unsigned int priority,
1826 const struct ofpact *ofpacts, size_t ofpacts_len)
1827 OVS_EXCLUDED(ofproto_mutex)
1829 const struct rule *rule;
1832 /* First do a cheap check whether the rule we're looking for already exists
1833 * with the actions that we want. If it does, then we're done. */
1834 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1835 rule = rule_from_cls_rule(classifier_find_match_exactly(
1836 &ofproto->tables[0].cls, match, priority));
1838 ovs_mutex_lock(&rule->mutex);
1839 must_add = !ofpacts_equal(rule->actions->ofpacts,
1840 rule->actions->ofpacts_len,
1841 ofpacts, ofpacts_len);
1842 ovs_mutex_unlock(&rule->mutex);
1846 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1848 /* If there's no such rule or the rule doesn't have the actions we want,
1849 * fall back to a executing a full flow mod. We can't optimize this at
1850 * all because we didn't take enough locks above to ensure that the flow
1851 * table didn't already change beneath us. */
1853 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1854 OFPFC_MODIFY_STRICT);
1858 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1859 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1860 * operation cannot be initiated now but may be retried later.
1862 * This is a helper function for in-band control and fail-open. */
1864 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1865 OVS_EXCLUDED(ofproto_mutex)
1867 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1870 /* Searches for a rule with matching criteria exactly equal to 'target' in
1871 * ofproto's table 0 and, if it finds one, deletes it.
1873 * This is a helper function for in-band control and fail-open. */
1875 ofproto_delete_flow(struct ofproto *ofproto,
1876 const struct match *target, unsigned int priority)
1877 OVS_EXCLUDED(ofproto_mutex)
1879 struct classifier *cls = &ofproto->tables[0].cls;
1882 /* First do a cheap check whether the rule we're looking for has already
1883 * been deleted. If so, then we're done. */
1884 ovs_rwlock_rdlock(&cls->rwlock);
1885 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1887 ovs_rwlock_unlock(&cls->rwlock);
1892 /* Fall back to a executing a full flow mod. We can't optimize this at all
1893 * because we didn't take enough locks above to ensure that the flow table
1894 * didn't already change beneath us. */
1895 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1896 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1899 /* Starts the process of deleting all of the flows from all of ofproto's flow
1900 * tables and then reintroducing the flows required by in-band control and
1901 * fail-open. The process will complete in a later call to ofproto_run(). */
1903 ofproto_flush_flows(struct ofproto *ofproto)
1905 COVERAGE_INC(ofproto_flush);
1906 ofproto->state = S_FLUSH;
1910 reinit_ports(struct ofproto *p)
1912 struct ofproto_port_dump dump;
1913 struct sset devnames;
1914 struct ofport *ofport;
1915 struct ofproto_port ofproto_port;
1916 const char *devname;
1918 COVERAGE_INC(ofproto_reinit_ports);
1920 sset_init(&devnames);
1921 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1922 sset_add(&devnames, netdev_get_name(ofport->netdev));
1924 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1925 sset_add(&devnames, ofproto_port.name);
1928 SSET_FOR_EACH (devname, &devnames) {
1929 update_port(p, devname);
1931 sset_destroy(&devnames);
1935 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1939 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1940 port_idx = port_idx ? port_idx : UINT16_MAX;
1942 if (port_idx >= ofproto->max_ports
1943 || bitmap_is_set(ofproto->ofp_port_ids, port_idx)) {
1944 uint16_t end_port_no = ofproto->alloc_port_no;
1946 /* Search for a free OpenFlow port number. We try not to
1947 * immediately reuse them to prevent problems due to old
1950 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1951 ofproto->alloc_port_no = 0;
1953 if (!bitmap_is_set(ofproto->ofp_port_ids,
1954 ofproto->alloc_port_no)) {
1955 port_idx = ofproto->alloc_port_no;
1958 if (ofproto->alloc_port_no == end_port_no) {
1963 bitmap_set1(ofproto->ofp_port_ids, port_idx);
1964 return u16_to_ofp(port_idx);
1968 dealloc_ofp_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
1970 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1971 bitmap_set0(ofproto->ofp_port_ids, ofp_to_u16(ofp_port));
1975 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1976 * pointer if the netdev cannot be opened. On success, also fills in
1978 static struct netdev *
1979 ofport_open(struct ofproto *ofproto,
1980 struct ofproto_port *ofproto_port,
1981 struct ofputil_phy_port *pp)
1983 enum netdev_flags flags;
1984 struct netdev *netdev;
1987 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1989 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1990 "cannot be opened (%s)",
1992 ofproto_port->name, ofproto_port->ofp_port,
1993 ofproto_port->name, ovs_strerror(error));
1997 if (ofproto_port->ofp_port == OFPP_NONE) {
1998 if (!strcmp(ofproto->name, ofproto_port->name)) {
1999 ofproto_port->ofp_port = OFPP_LOCAL;
2001 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2002 ofproto_port->name);
2005 pp->port_no = ofproto_port->ofp_port;
2006 netdev_get_etheraddr(netdev, pp->hw_addr);
2007 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2008 netdev_get_flags(netdev, &flags);
2009 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2010 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2011 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2012 &pp->supported, &pp->peer);
2013 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2014 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2019 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2020 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2023 ofport_equal(const struct ofputil_phy_port *a,
2024 const struct ofputil_phy_port *b)
2026 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2027 && a->state == b->state
2028 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2029 && a->curr == b->curr
2030 && a->advertised == b->advertised
2031 && a->supported == b->supported
2032 && a->peer == b->peer
2033 && a->curr_speed == b->curr_speed
2034 && a->max_speed == b->max_speed);
2037 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2038 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2039 * one with the same name or port number). */
2041 ofport_install(struct ofproto *p,
2042 struct netdev *netdev, const struct ofputil_phy_port *pp)
2044 const char *netdev_name = netdev_get_name(netdev);
2045 struct ofport *ofport;
2048 /* Create ofport. */
2049 ofport = p->ofproto_class->port_alloc();
2054 ofport->ofproto = p;
2055 ofport->netdev = netdev;
2056 ofport->change_seq = netdev_change_seq(netdev);
2058 ofport->ofp_port = pp->port_no;
2059 ofport->created = time_msec();
2061 /* Add port to 'p'. */
2062 hmap_insert(&p->ports, &ofport->hmap_node,
2063 hash_ofp_port(ofport->ofp_port));
2064 shash_add(&p->port_by_name, netdev_name, ofport);
2066 update_mtu(p, ofport);
2068 /* Let the ofproto_class initialize its private data. */
2069 error = p->ofproto_class->port_construct(ofport);
2073 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2077 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2078 p->name, netdev_name, ovs_strerror(error));
2080 ofport_destroy__(ofport);
2082 netdev_close(netdev);
2086 /* Removes 'ofport' from 'p' and destroys it. */
2088 ofport_remove(struct ofport *ofport)
2090 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2092 ofport_destroy(ofport);
2095 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2098 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2100 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2102 ofport_remove(port);
2106 /* Updates 'port' with new 'pp' description.
2108 * Does not handle a name or port number change. The caller must implement
2109 * such a change as a delete followed by an add. */
2111 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2113 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2114 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2115 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2116 port->pp.state = pp->state;
2117 port->pp.curr = pp->curr;
2118 port->pp.advertised = pp->advertised;
2119 port->pp.supported = pp->supported;
2120 port->pp.peer = pp->peer;
2121 port->pp.curr_speed = pp->curr_speed;
2122 port->pp.max_speed = pp->max_speed;
2124 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2127 /* Update OpenFlow 'state' in 'port' and notify controller. */
2129 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2131 if (port->pp.state != state) {
2132 port->pp.state = state;
2133 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2139 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2141 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2143 if (port->ofproto->ofproto_class->set_realdev) {
2144 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2146 if (port->ofproto->ofproto_class->set_stp_port) {
2147 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2149 if (port->ofproto->ofproto_class->set_cfm) {
2150 port->ofproto->ofproto_class->set_cfm(port, NULL);
2152 if (port->ofproto->ofproto_class->bundle_remove) {
2153 port->ofproto->ofproto_class->bundle_remove(port);
2159 ofport_destroy__(struct ofport *port)
2161 struct ofproto *ofproto = port->ofproto;
2162 const char *name = netdev_get_name(port->netdev);
2164 hmap_remove(&ofproto->ports, &port->hmap_node);
2165 shash_delete(&ofproto->port_by_name,
2166 shash_find(&ofproto->port_by_name, name));
2168 netdev_close(port->netdev);
2169 ofproto->ofproto_class->port_dealloc(port);
2173 ofport_destroy(struct ofport *port)
2176 dealloc_ofp_port(port->ofproto, port->ofp_port);
2177 port->ofproto->ofproto_class->port_destruct(port);
2178 ofport_destroy__(port);
2183 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2185 struct ofport *port;
2187 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2189 if (port->ofp_port == ofp_port) {
2197 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2199 struct ofproto *ofproto = port->ofproto;
2202 if (ofproto->ofproto_class->port_get_stats) {
2203 error = ofproto->ofproto_class->port_get_stats(port, stats);
2212 update_port(struct ofproto *ofproto, const char *name)
2214 struct ofproto_port ofproto_port;
2215 struct ofputil_phy_port pp;
2216 struct netdev *netdev;
2217 struct ofport *port;
2219 COVERAGE_INC(ofproto_update_port);
2221 /* Fetch 'name''s location and properties from the datapath. */
2222 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2223 ? ofport_open(ofproto, &ofproto_port, &pp)
2227 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2228 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2229 struct netdev *old_netdev = port->netdev;
2231 /* 'name' hasn't changed location. Any properties changed? */
2232 if (!ofport_equal(&port->pp, &pp)) {
2233 ofport_modified(port, &pp);
2236 update_mtu(ofproto, port);
2238 /* Install the newly opened netdev in case it has changed.
2239 * Don't close the old netdev yet in case port_modified has to
2240 * remove a retained reference to it.*/
2241 port->netdev = netdev;
2242 port->change_seq = netdev_change_seq(netdev);
2244 if (port->ofproto->ofproto_class->port_modified) {
2245 port->ofproto->ofproto_class->port_modified(port);
2248 netdev_close(old_netdev);
2250 /* If 'port' is nonnull then its name differs from 'name' and thus
2251 * we should delete it. If we think there's a port named 'name'
2252 * then its port number must be wrong now so delete it too. */
2254 ofport_remove(port);
2256 ofport_remove_with_name(ofproto, name);
2257 ofport_install(ofproto, netdev, &pp);
2260 /* Any port named 'name' is gone now. */
2261 ofport_remove_with_name(ofproto, name);
2263 ofproto_port_destroy(&ofproto_port);
2267 init_ports(struct ofproto *p)
2269 struct ofproto_port_dump dump;
2270 struct ofproto_port ofproto_port;
2271 struct shash_node *node, *next;
2273 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2274 const char *name = ofproto_port.name;
2276 if (shash_find(&p->port_by_name, name)) {
2277 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2280 struct ofputil_phy_port pp;
2281 struct netdev *netdev;
2283 /* Check if an OpenFlow port number had been requested. */
2284 node = shash_find(&init_ofp_ports, name);
2286 const struct iface_hint *iface_hint = node->data;
2287 simap_put(&p->ofp_requests, name,
2288 ofp_to_u16(iface_hint->ofp_port));
2291 netdev = ofport_open(p, &ofproto_port, &pp);
2293 ofport_install(p, netdev, &pp);
2294 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2295 p->alloc_port_no = MAX(p->alloc_port_no,
2296 ofp_to_u16(ofproto_port.ofp_port));
2302 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2303 struct iface_hint *iface_hint = node->data;
2305 if (!strcmp(iface_hint->br_name, p->name)) {
2306 free(iface_hint->br_name);
2307 free(iface_hint->br_type);
2309 shash_delete(&init_ofp_ports, node);
2316 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2317 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2319 find_min_mtu(struct ofproto *p)
2321 struct ofport *ofport;
2324 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2325 struct netdev *netdev = ofport->netdev;
2328 /* Skip any internal ports, since that's what we're trying to
2330 if (!strcmp(netdev_get_type(netdev), "internal")) {
2334 if (netdev_get_mtu(netdev, &dev_mtu)) {
2337 if (!mtu || dev_mtu < mtu) {
2342 return mtu ? mtu: ETH_PAYLOAD_MAX;
2345 /* Update MTU of all datapath devices on 'p' to the minimum of the
2346 * non-datapath ports in event of 'port' added or changed. */
2348 update_mtu(struct ofproto *p, struct ofport *port)
2350 struct ofport *ofport;
2351 struct netdev *netdev = port->netdev;
2352 int dev_mtu, old_min;
2354 if (netdev_get_mtu(netdev, &dev_mtu)) {
2358 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2359 if (dev_mtu > p->min_mtu) {
2360 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2361 dev_mtu = p->min_mtu;
2364 port->mtu = dev_mtu;
2368 /* For non-internal port find new min mtu. */
2369 old_min = p->min_mtu;
2370 port->mtu = dev_mtu;
2371 p->min_mtu = find_min_mtu(p);
2372 if (p->min_mtu == old_min) {
2376 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2377 struct netdev *netdev = ofport->netdev;
2379 if (!strcmp(netdev_get_type(netdev), "internal")) {
2380 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2381 ofport->mtu = p->min_mtu;
2388 ofproto_rule_ref(struct rule *rule)
2393 atomic_add(&rule->ref_count, 1, &orig);
2394 ovs_assert(orig != 0);
2399 ofproto_rule_unref(struct rule *rule)
2404 atomic_sub(&rule->ref_count, 1, &orig);
2406 rule->ofproto->ofproto_class->rule_destruct(rule);
2407 ofproto_rule_destroy__(rule);
2409 ovs_assert(orig != 0);
2414 struct rule_actions *
2415 rule_get_actions(const struct rule *rule)
2416 OVS_EXCLUDED(rule->mutex)
2418 struct rule_actions *actions;
2420 ovs_mutex_lock(&rule->mutex);
2421 actions = rule_get_actions__(rule);
2422 ovs_mutex_unlock(&rule->mutex);
2427 struct rule_actions *
2428 rule_get_actions__(const struct rule *rule)
2429 OVS_REQUIRES(rule->mutex)
2431 rule_actions_ref(rule->actions);
2432 return rule->actions;
2436 ofproto_rule_destroy__(struct rule *rule)
2437 OVS_NO_THREAD_SAFETY_ANALYSIS
2439 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2440 rule_actions_unref(rule->actions);
2441 ovs_mutex_destroy(&rule->mutex);
2442 rule->ofproto->ofproto_class->rule_dealloc(rule);
2445 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2446 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2447 struct rule_actions *
2448 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2450 struct rule_actions *actions;
2452 actions = xmalloc(sizeof *actions);
2453 atomic_init(&actions->ref_count, 1);
2454 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2455 actions->ofpacts_len = ofpacts_len;
2456 actions->meter_id = ofpacts_get_meter(ofpacts, ofpacts_len);
2460 /* Increments 'actions''s ref_count. */
2462 rule_actions_ref(struct rule_actions *actions)
2467 atomic_add(&actions->ref_count, 1, &orig);
2468 ovs_assert(orig != 0);
2472 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2475 rule_actions_unref(struct rule_actions *actions)
2480 atomic_sub(&actions->ref_count, 1, &orig);
2484 ovs_assert(orig != 0);
2489 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2490 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2492 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2493 OVS_REQUIRES(ofproto_mutex)
2495 return (port == OFPP_ANY
2496 || ofpacts_output_to_port(rule->actions->ofpacts,
2497 rule->actions->ofpacts_len, port));
2500 /* Returns true if 'rule' has group and equals group_id. */
2502 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2503 OVS_REQUIRES(ofproto_mutex)
2505 return (group_id == OFPG11_ANY
2506 || ofpacts_output_to_group(rule->actions->ofpacts,
2507 rule->actions->ofpacts_len, group_id));
2510 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2511 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2513 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2514 OVS_REQUIRES(ofproto_mutex)
2516 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2521 case OFOPERATION_ADD:
2522 case OFOPERATION_DELETE:
2525 case OFOPERATION_MODIFY:
2526 case OFOPERATION_REPLACE:
2527 return ofpacts_output_to_port(op->actions->ofpacts,
2528 op->actions->ofpacts_len, out_port);
2535 rule_execute_destroy(struct rule_execute *e)
2537 ofproto_rule_unref(e->rule);
2538 list_remove(&e->list_node);
2542 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2543 * by passing them to the ofproto provider. */
2545 run_rule_executes(struct ofproto *ofproto)
2546 OVS_EXCLUDED(ofproto_mutex)
2548 struct rule_execute *e, *next;
2549 struct list executes;
2551 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2552 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2553 union flow_in_port in_port_;
2556 in_port_.ofp_port = e->in_port;
2557 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2558 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2560 rule_execute_destroy(e);
2564 /* Destroys and discards all "rule_execute" operations queued up in
2565 * ofproto->rule_executes. */
2567 destroy_rule_executes(struct ofproto *ofproto)
2569 struct rule_execute *e, *next;
2570 struct list executes;
2572 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2573 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2574 ofpbuf_delete(e->packet);
2575 rule_execute_destroy(e);
2579 /* Returns true if 'rule' should be hidden from the controller.
2581 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2582 * (e.g. by in-band control) and are intentionally hidden from the
2585 ofproto_rule_is_hidden(const struct rule *rule)
2587 return rule->cr.priority > UINT16_MAX;
2590 static enum oftable_flags
2591 rule_get_flags(const struct rule *rule)
2593 return rule->ofproto->tables[rule->table_id].flags;
2597 rule_is_modifiable(const struct rule *rule)
2599 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2603 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2605 ofconn_send_reply(ofconn, make_echo_reply(oh));
2610 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2612 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2613 struct ofputil_switch_features features;
2614 struct ofport *port;
2620 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2622 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2624 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2625 * if present, are always at the end.) */
2626 n_tables = ofproto->n_tables;
2627 for (i = 0; i < ofproto->n_tables; i++) {
2628 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2634 features.datapath_id = ofproto->datapath_id;
2635 features.n_buffers = pktbuf_capacity();
2636 features.n_tables = n_tables;
2637 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2638 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2640 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2642 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2643 features.auxiliary_id = 0;
2644 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2646 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2647 ofputil_put_switch_features_port(&port->pp, b);
2650 ofconn_send_reply(ofconn, b);
2655 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2657 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2658 struct ofp_switch_config *osc;
2659 enum ofp_config_flags flags;
2663 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2664 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2665 flags = ofproto->frag_handling;
2666 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2667 if (oh->version < OFP13_VERSION
2668 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2669 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2671 osc->flags = htons(flags);
2672 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2673 ofconn_send_reply(ofconn, buf);
2679 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2681 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2682 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2683 uint16_t flags = ntohs(osc->flags);
2685 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2686 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2687 enum ofp_config_flags cur = ofproto->frag_handling;
2688 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2690 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2692 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2693 ofproto->frag_handling = next;
2695 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2697 ofputil_frag_handling_to_string(next));
2701 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2702 ofconn_set_invalid_ttl_to_controller(ofconn,
2703 (oh->version < OFP13_VERSION
2704 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2706 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2711 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2712 * error message code for the caller to propagate upward. Otherwise, returns
2715 * The log message mentions 'msg_type'. */
2717 reject_slave_controller(struct ofconn *ofconn)
2719 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2720 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2721 return OFPERR_OFPBRC_EPERM;
2727 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2728 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2729 * 'flow' may be temporarily modified, but is restored at return.
2732 ofproto_check_ofpacts(struct ofproto *ofproto,
2733 const struct ofpact ofpacts[], size_t ofpacts_len,
2734 struct flow *flow, uint8_t table_id)
2739 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2740 u16_to_ofp(ofproto->max_ports), table_id);
2745 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2746 if (mid && ofproto_get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2747 return OFPERR_OFPMMFC_INVALID_METER;
2753 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2755 struct ofproto *p = ofconn_get_ofproto(ofconn);
2756 struct ofputil_packet_out po;
2757 struct ofpbuf *payload;
2758 uint64_t ofpacts_stub[1024 / 8];
2759 struct ofpbuf ofpacts;
2761 union flow_in_port in_port_;
2764 COVERAGE_INC(ofproto_packet_out);
2766 error = reject_slave_controller(ofconn);
2771 /* Decode message. */
2772 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2773 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2775 goto exit_free_ofpacts;
2777 if (ofp_to_u16(po.in_port) >= p->max_ports
2778 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2779 error = OFPERR_OFPBRC_BAD_PORT;
2780 goto exit_free_ofpacts;
2785 if (po.buffer_id != UINT32_MAX) {
2786 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2787 if (error || !payload) {
2788 goto exit_free_ofpacts;
2791 /* Ensure that the L3 header is 32-bit aligned. */
2792 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2795 /* Verify actions against packet, then send packet if successful. */
2796 in_port_.ofp_port = po.in_port;
2797 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2798 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2800 error = p->ofproto_class->packet_out(p, payload, &flow,
2801 po.ofpacts, po.ofpacts_len);
2803 ofpbuf_delete(payload);
2806 ofpbuf_uninit(&ofpacts);
2812 update_port_config(struct ofport *port,
2813 enum ofputil_port_config config,
2814 enum ofputil_port_config mask)
2816 enum ofputil_port_config old_config = port->pp.config;
2817 enum ofputil_port_config toggle;
2819 toggle = (config ^ port->pp.config) & mask;
2820 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2821 if (config & OFPUTIL_PC_PORT_DOWN) {
2822 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2824 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2826 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2829 port->pp.config ^= toggle;
2830 if (port->pp.config != old_config) {
2831 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2836 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2838 struct ofproto *p = ofconn_get_ofproto(ofconn);
2839 struct ofputil_port_mod pm;
2840 struct ofport *port;
2843 error = reject_slave_controller(ofconn);
2848 error = ofputil_decode_port_mod(oh, &pm);
2853 port = ofproto_get_port(p, pm.port_no);
2855 return OFPERR_OFPPMFC_BAD_PORT;
2856 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2857 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2859 update_port_config(port, pm.config, pm.mask);
2861 netdev_set_advertisements(port->netdev, pm.advertise);
2868 handle_desc_stats_request(struct ofconn *ofconn,
2869 const struct ofp_header *request)
2871 static const char *default_mfr_desc = "Nicira, Inc.";
2872 static const char *default_hw_desc = "Open vSwitch";
2873 static const char *default_sw_desc = VERSION;
2874 static const char *default_serial_desc = "None";
2875 static const char *default_dp_desc = "None";
2877 struct ofproto *p = ofconn_get_ofproto(ofconn);
2878 struct ofp_desc_stats *ods;
2881 msg = ofpraw_alloc_stats_reply(request, 0);
2882 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2883 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2884 sizeof ods->mfr_desc);
2885 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2886 sizeof ods->hw_desc);
2887 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2888 sizeof ods->sw_desc);
2889 ovs_strlcpy(ods->serial_num,
2890 p->serial_desc ? p->serial_desc : default_serial_desc,
2891 sizeof ods->serial_num);
2892 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2893 sizeof ods->dp_desc);
2894 ofconn_send_reply(ofconn, msg);
2900 handle_table_stats_request(struct ofconn *ofconn,
2901 const struct ofp_header *request)
2903 struct ofproto *p = ofconn_get_ofproto(ofconn);
2904 struct ofp12_table_stats *ots;
2909 /* Set up default values.
2911 * ofp12_table_stats is used as a generic structure as
2912 * it is able to hold all the fields for ofp10_table_stats
2913 * and ofp11_table_stats (and of course itself).
2915 ots = xcalloc(p->n_tables, sizeof *ots);
2916 for (i = 0; i < p->n_tables; i++) {
2917 ots[i].table_id = i;
2918 sprintf(ots[i].name, "table%zu", i);
2919 ots[i].match = htonll(OFPXMT12_MASK);
2920 ots[i].wildcards = htonll(OFPXMT12_MASK);
2921 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2922 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2923 ots[i].write_setfields = htonll(OFPXMT12_MASK);
2924 ots[i].apply_setfields = htonll(OFPXMT12_MASK);
2925 ots[i].metadata_match = htonll(UINT64_MAX);
2926 ots[i].metadata_write = htonll(UINT64_MAX);
2927 ots[i].instructions = htonl(OFPIT11_ALL);
2928 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2929 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2930 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
2931 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2932 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
2935 p->ofproto_class->get_tables(p, ots);
2937 /* Post-process the tables, dropping hidden tables. */
2938 n_tables = p->n_tables;
2939 for (i = 0; i < p->n_tables; i++) {
2940 const struct oftable *table = &p->tables[i];
2942 if (table->flags & OFTABLE_HIDDEN) {
2948 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2951 if (table->max_flows < ntohl(ots[i].max_entries)) {
2952 ots[i].max_entries = htonl(table->max_flows);
2956 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
2957 ofconn_send_reply(ofconn, msg);
2965 append_port_stat(struct ofport *port, struct list *replies)
2967 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
2969 calc_duration(port->created, time_msec(),
2970 &ops.duration_sec, &ops.duration_nsec);
2972 /* Intentionally ignore return value, since errors will set
2973 * 'stats' to all-1s, which is correct for OpenFlow, and
2974 * netdev_get_stats() will log errors. */
2975 ofproto_port_get_stats(port, &ops.stats);
2977 ofputil_append_port_stat(replies, &ops);
2981 handle_port_stats_request(struct ofconn *ofconn,
2982 const struct ofp_header *request)
2984 struct ofproto *p = ofconn_get_ofproto(ofconn);
2985 struct ofport *port;
2986 struct list replies;
2990 error = ofputil_decode_port_stats_request(request, &port_no);
2995 ofpmp_init(&replies, request);
2996 if (port_no != OFPP_ANY) {
2997 port = ofproto_get_port(p, port_no);
2999 append_port_stat(port, &replies);
3002 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3003 append_port_stat(port, &replies);
3007 ofconn_send_replies(ofconn, &replies);
3012 handle_port_desc_stats_request(struct ofconn *ofconn,
3013 const struct ofp_header *request)
3015 struct ofproto *p = ofconn_get_ofproto(ofconn);
3016 enum ofp_version version;
3017 struct ofport *port;
3018 struct list replies;
3020 ofpmp_init(&replies, request);
3022 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3023 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3024 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3027 ofconn_send_replies(ofconn, &replies);
3032 hash_cookie(ovs_be64 cookie)
3034 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3035 (OVS_FORCE uint64_t)cookie);
3039 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3040 OVS_REQUIRES(ofproto_mutex)
3042 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3043 hash_cookie(rule->flow_cookie));
3047 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3048 OVS_REQUIRES(ofproto_mutex)
3050 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3054 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3055 ovs_be64 new_cookie)
3056 OVS_REQUIRES(ofproto_mutex)
3058 if (new_cookie != rule->flow_cookie) {
3059 cookies_remove(ofproto, rule);
3061 ovs_mutex_lock(&rule->mutex);
3062 rule->flow_cookie = new_cookie;
3063 ovs_mutex_unlock(&rule->mutex);
3065 cookies_insert(ofproto, rule);
3070 calc_duration(long long int start, long long int now,
3071 uint32_t *sec, uint32_t *nsec)
3073 long long int msecs = now - start;
3074 *sec = msecs / 1000;
3075 *nsec = (msecs % 1000) * (1000 * 1000);
3078 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3079 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3081 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3083 return (table_id == 0xff || table_id < ofproto->n_tables
3085 : OFPERR_OFPBRC_BAD_TABLE_ID);
3089 static struct oftable *
3090 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3092 struct oftable *table;
3094 for (table = &ofproto->tables[table_id];
3095 table < &ofproto->tables[ofproto->n_tables];
3097 if (!(table->flags & OFTABLE_HIDDEN)) {
3105 static struct oftable *
3106 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3108 if (table_id == 0xff) {
3109 return next_visible_table(ofproto, 0);
3110 } else if (table_id < ofproto->n_tables) {
3111 return &ofproto->tables[table_id];
3117 static struct oftable *
3118 next_matching_table(const struct ofproto *ofproto,
3119 const struct oftable *table, uint8_t table_id)
3121 return (table_id == 0xff
3122 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3126 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3128 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3129 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3131 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3132 * only once, for that table. (This can be used to access tables marked
3135 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3136 * entered at all. (Perhaps you should have validated TABLE_ID with
3137 * check_table_id().)
3139 * All parameters are evaluated multiple times.
3141 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3142 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3144 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3146 /* Initializes 'criteria' in a straightforward way based on the other
3149 * For "loose" matching, the 'priority' parameter is unimportant and may be
3152 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3153 const struct match *match, unsigned int priority,
3154 ovs_be64 cookie, ovs_be64 cookie_mask,
3155 ofp_port_t out_port, uint32_t out_group)
3157 criteria->table_id = table_id;
3158 cls_rule_init(&criteria->cr, match, priority);
3159 criteria->cookie = cookie;
3160 criteria->cookie_mask = cookie_mask;
3161 criteria->out_port = out_port;
3162 criteria->out_group = out_group;
3166 rule_criteria_destroy(struct rule_criteria *criteria)
3168 cls_rule_destroy(&criteria->cr);
3172 rule_collection_init(struct rule_collection *rules)
3174 rules->rules = rules->stub;
3176 rules->capacity = ARRAY_SIZE(rules->stub);
3180 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3182 if (rules->n >= rules->capacity) {
3183 size_t old_size, new_size;
3185 old_size = rules->capacity * sizeof *rules->rules;
3186 rules->capacity *= 2;
3187 new_size = rules->capacity * sizeof *rules->rules;
3189 if (rules->rules == rules->stub) {
3190 rules->rules = xmalloc(new_size);
3191 memcpy(rules->rules, rules->stub, old_size);
3193 rules->rules = xrealloc(rules->rules, new_size);
3197 rules->rules[rules->n++] = rule;
3201 rule_collection_ref(struct rule_collection *rules)
3202 OVS_REQUIRES(ofproto_mutex)
3206 for (i = 0; i < rules->n; i++) {
3207 ofproto_rule_ref(rules->rules[i]);
3212 rule_collection_unref(struct rule_collection *rules)
3216 for (i = 0; i < rules->n; i++) {
3217 ofproto_rule_unref(rules->rules[i]);
3222 rule_collection_destroy(struct rule_collection *rules)
3224 if (rules->rules != rules->stub) {
3230 collect_rule(struct rule *rule, const struct rule_criteria *c,
3231 struct rule_collection *rules)
3232 OVS_REQUIRES(ofproto_mutex)
3234 if (ofproto_rule_is_hidden(rule)) {
3236 } else if (rule->pending) {
3237 return OFPROTO_POSTPONE;
3239 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3240 && ofproto_rule_has_out_port(rule, c->out_port)
3241 && ofproto_rule_has_out_group(rule, c->out_group)
3242 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3243 rule_collection_add(rules, rule);
3249 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3250 * on classifiers rules are done in the "loose" way required for OpenFlow
3251 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3254 * Hidden rules are always omitted.
3256 * Returns 0 on success, otherwise an OpenFlow error code. */
3258 collect_rules_loose(struct ofproto *ofproto,
3259 const struct rule_criteria *criteria,
3260 struct rule_collection *rules)
3261 OVS_REQUIRES(ofproto_mutex)
3263 struct oftable *table;
3266 rule_collection_init(rules);
3268 error = check_table_id(ofproto, criteria->table_id);
3273 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3276 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3277 hash_cookie(criteria->cookie),
3278 &ofproto->cookies) {
3279 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3280 error = collect_rule(rule, criteria, rules);
3287 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3288 struct cls_cursor cursor;
3291 ovs_rwlock_rdlock(&table->cls.rwlock);
3292 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3293 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3294 error = collect_rule(rule, criteria, rules);
3299 ovs_rwlock_unlock(&table->cls.rwlock);
3305 rule_collection_destroy(rules);
3310 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3311 * on classifiers rules are done in the "strict" way required for OpenFlow
3312 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3313 * rules on list 'rules'.
3315 * Hidden rules are always omitted.
3317 * Returns 0 on success, otherwise an OpenFlow error code. */
3319 collect_rules_strict(struct ofproto *ofproto,
3320 const struct rule_criteria *criteria,
3321 struct rule_collection *rules)
3322 OVS_REQUIRES(ofproto_mutex)
3324 struct oftable *table;
3327 rule_collection_init(rules);
3329 error = check_table_id(ofproto, criteria->table_id);
3334 if (criteria->cookie_mask == htonll(UINT64_MAX)) {
3337 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3338 hash_cookie(criteria->cookie),
3339 &ofproto->cookies) {
3340 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3341 error = collect_rule(rule, criteria, rules);
3348 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3351 ovs_rwlock_rdlock(&table->cls.rwlock);
3352 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3353 &table->cls, &criteria->cr));
3354 ovs_rwlock_unlock(&table->cls.rwlock);
3356 error = collect_rule(rule, criteria, rules);
3366 rule_collection_destroy(rules);
3371 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3372 * forced into the range of a uint16_t. */
3374 age_secs(long long int age_ms)
3376 return (age_ms < 0 ? 0
3377 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3378 : (unsigned int) age_ms / 1000);
3382 handle_flow_stats_request(struct ofconn *ofconn,
3383 const struct ofp_header *request)
3384 OVS_EXCLUDED(ofproto_mutex)
3386 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3387 struct ofputil_flow_stats_request fsr;
3388 struct rule_criteria criteria;
3389 struct rule_collection rules;
3390 struct list replies;
3394 error = ofputil_decode_flow_stats_request(&fsr, request);
3399 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3400 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3402 ovs_mutex_lock(&ofproto_mutex);
3403 error = collect_rules_loose(ofproto, &criteria, &rules);
3404 rule_criteria_destroy(&criteria);
3406 rule_collection_ref(&rules);
3408 ovs_mutex_unlock(&ofproto_mutex);
3414 ofpmp_init(&replies, request);
3415 for (i = 0; i < rules.n; i++) {
3416 struct rule *rule = rules.rules[i];
3417 long long int now = time_msec();
3418 struct ofputil_flow_stats fs;
3419 long long int created, used, modified;
3420 struct rule_actions *actions;
3421 enum ofputil_flow_mod_flags flags;
3423 ovs_mutex_lock(&rule->mutex);
3424 fs.cookie = rule->flow_cookie;
3425 fs.idle_timeout = rule->idle_timeout;
3426 fs.hard_timeout = rule->hard_timeout;
3427 created = rule->created;
3429 modified = rule->modified;
3430 actions = rule_get_actions__(rule);
3431 flags = rule->flags;
3432 ovs_mutex_unlock(&rule->mutex);
3434 minimatch_expand(&rule->cr.match, &fs.match);
3435 fs.table_id = rule->table_id;
3436 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3437 fs.priority = rule->cr.priority;
3438 fs.idle_age = age_secs(now - used);
3439 fs.hard_age = age_secs(now - modified);
3440 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3442 fs.ofpacts = actions->ofpacts;
3443 fs.ofpacts_len = actions->ofpacts_len;
3446 ofputil_append_flow_stats_reply(&fs, &replies);
3448 rule_actions_unref(actions);
3451 rule_collection_unref(&rules);
3452 rule_collection_destroy(&rules);
3454 ofconn_send_replies(ofconn, &replies);
3460 flow_stats_ds(struct rule *rule, struct ds *results)
3462 uint64_t packet_count, byte_count;
3463 struct rule_actions *actions;
3464 long long int created;
3466 rule->ofproto->ofproto_class->rule_get_stats(rule,
3467 &packet_count, &byte_count);
3469 ovs_mutex_lock(&rule->mutex);
3470 actions = rule_get_actions__(rule);
3471 created = rule->created;
3472 ovs_mutex_unlock(&rule->mutex);
3474 if (rule->table_id != 0) {
3475 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3477 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3478 ds_put_format(results, "priority=%u, ", rule->cr.priority);
3479 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3480 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3481 cls_rule_format(&rule->cr, results);
3482 ds_put_char(results, ',');
3484 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3486 ds_put_cstr(results, "\n");
3488 rule_actions_unref(actions);
3491 /* Adds a pretty-printed description of all flows to 'results', including
3492 * hidden flows (e.g., set up by in-band control). */
3494 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3496 struct oftable *table;
3498 OFPROTO_FOR_EACH_TABLE (table, p) {
3499 struct cls_cursor cursor;
3502 ovs_rwlock_rdlock(&table->cls.rwlock);
3503 cls_cursor_init(&cursor, &table->cls, NULL);
3504 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3505 flow_stats_ds(rule, results);
3507 ovs_rwlock_unlock(&table->cls.rwlock);
3511 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3512 * '*engine_type' and '*engine_id', respectively. */
3514 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3515 uint8_t *engine_type, uint8_t *engine_id)
3517 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3520 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3521 * true if the port's CFM status was successfully stored into '*status'.
3522 * Returns false if the port did not have CFM configured, in which case
3523 * '*status' is indeterminate.
3525 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3527 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3528 struct ofproto_cfm_status *status)
3530 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3532 && ofproto->ofproto_class->get_cfm_status
3533 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3537 handle_aggregate_stats_request(struct ofconn *ofconn,
3538 const struct ofp_header *oh)
3539 OVS_EXCLUDED(ofproto_mutex)
3541 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3542 struct ofputil_flow_stats_request request;
3543 struct ofputil_aggregate_stats stats;
3544 bool unknown_packets, unknown_bytes;
3545 struct rule_criteria criteria;
3546 struct rule_collection rules;
3547 struct ofpbuf *reply;
3551 error = ofputil_decode_flow_stats_request(&request, oh);
3556 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3557 request.cookie, request.cookie_mask,
3558 request.out_port, request.out_group);
3560 ovs_mutex_lock(&ofproto_mutex);
3561 error = collect_rules_loose(ofproto, &criteria, &rules);
3562 rule_criteria_destroy(&criteria);
3564 rule_collection_ref(&rules);
3566 ovs_mutex_unlock(&ofproto_mutex);
3572 memset(&stats, 0, sizeof stats);
3573 unknown_packets = unknown_bytes = false;
3574 for (i = 0; i < rules.n; i++) {
3575 struct rule *rule = rules.rules[i];
3576 uint64_t packet_count;
3577 uint64_t byte_count;
3579 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3582 if (packet_count == UINT64_MAX) {
3583 unknown_packets = true;
3585 stats.packet_count += packet_count;
3588 if (byte_count == UINT64_MAX) {
3589 unknown_bytes = true;
3591 stats.byte_count += byte_count;
3596 if (unknown_packets) {
3597 stats.packet_count = UINT64_MAX;
3599 if (unknown_bytes) {
3600 stats.byte_count = UINT64_MAX;
3603 rule_collection_unref(&rules);
3604 rule_collection_destroy(&rules);
3606 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3607 ofconn_send_reply(ofconn, reply);
3612 struct queue_stats_cbdata {
3613 struct ofport *ofport;
3614 struct list replies;
3619 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3620 const struct netdev_queue_stats *stats)
3622 struct ofputil_queue_stats oqs;
3624 oqs.port_no = cbdata->ofport->pp.port_no;
3625 oqs.queue_id = queue_id;
3626 oqs.tx_bytes = stats->tx_bytes;
3627 oqs.tx_packets = stats->tx_packets;
3628 oqs.tx_errors = stats->tx_errors;
3629 if (stats->created != LLONG_MIN) {
3630 calc_duration(stats->created, cbdata->now,
3631 &oqs.duration_sec, &oqs.duration_nsec);
3633 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3635 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3639 handle_queue_stats_dump_cb(uint32_t queue_id,
3640 struct netdev_queue_stats *stats,
3643 struct queue_stats_cbdata *cbdata = cbdata_;
3645 put_queue_stats(cbdata, queue_id, stats);
3649 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3650 struct queue_stats_cbdata *cbdata)
3652 cbdata->ofport = port;
3653 if (queue_id == OFPQ_ALL) {
3654 netdev_dump_queue_stats(port->netdev,
3655 handle_queue_stats_dump_cb, cbdata);
3657 struct netdev_queue_stats stats;
3659 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3660 put_queue_stats(cbdata, queue_id, &stats);
3662 return OFPERR_OFPQOFC_BAD_QUEUE;
3669 handle_queue_stats_request(struct ofconn *ofconn,
3670 const struct ofp_header *rq)
3672 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3673 struct queue_stats_cbdata cbdata;
3674 struct ofport *port;
3676 struct ofputil_queue_stats_request oqsr;
3678 COVERAGE_INC(ofproto_queue_req);
3680 ofpmp_init(&cbdata.replies, rq);
3681 cbdata.now = time_msec();
3683 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3688 if (oqsr.port_no == OFPP_ANY) {
3689 error = OFPERR_OFPQOFC_BAD_QUEUE;
3690 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3691 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3696 port = ofproto_get_port(ofproto, oqsr.port_no);
3698 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3699 : OFPERR_OFPQOFC_BAD_PORT);
3702 ofconn_send_replies(ofconn, &cbdata.replies);
3704 ofpbuf_list_delete(&cbdata.replies);
3711 is_flow_deletion_pending(const struct ofproto *ofproto,
3712 const struct cls_rule *cls_rule,
3714 OVS_REQUIRES(ofproto_mutex)
3716 if (!hmap_is_empty(&ofproto->deletions)) {
3717 struct ofoperation *op;
3719 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3720 cls_rule_hash(cls_rule, table_id),
3721 &ofproto->deletions) {
3722 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3732 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3733 OVS_REQUIRES(ofproto_mutex)
3734 OVS_NO_THREAD_SAFETY_ANALYSIS
3736 return classifier_count(&table->cls) + extra_space > table->max_flows;
3740 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3741 unsigned int extra_space)
3742 OVS_REQUIRES(ofproto_mutex)
3744 while (should_evict_a_rule(table, extra_space)) {
3747 if (!choose_rule_to_evict(table, &rule)) {
3748 return OFPERR_OFPFMFC_TABLE_FULL;
3749 } else if (rule->pending) {
3750 return OFPROTO_POSTPONE;
3752 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3753 delete_flow__(rule, group, OFPRR_EVICTION);
3754 ofopgroup_submit(group);
3761 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3762 * in which no matching flow already exists in the flow table.
3764 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3765 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3766 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3767 * initiated now but may be retried later.
3769 * The caller retains ownership of 'fm->ofpacts'.
3771 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3774 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3775 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3776 OVS_REQUIRES(ofproto_mutex)
3778 struct oftable *table;
3779 struct ofopgroup *group;
3785 error = check_table_id(ofproto, fm->table_id);
3791 if (fm->table_id == 0xff) {
3792 if (ofproto->ofproto_class->rule_choose_table) {
3793 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3799 ovs_assert(table_id < ofproto->n_tables);
3803 } else if (fm->table_id < ofproto->n_tables) {
3804 table_id = fm->table_id;
3806 return OFPERR_OFPBRC_BAD_TABLE_ID;
3809 table = &ofproto->tables[table_id];
3811 if (table->flags & OFTABLE_READONLY) {
3812 return OFPERR_OFPBRC_EPERM;
3815 cls_rule_init(&cr, &fm->match, fm->priority);
3817 /* Transform "add" into "modify" if there's an existing identical flow. */
3818 ovs_rwlock_rdlock(&table->cls.rwlock);
3819 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3820 ovs_rwlock_unlock(&table->cls.rwlock);
3822 cls_rule_destroy(&cr);
3823 if (!rule_is_modifiable(rule)) {
3824 return OFPERR_OFPBRC_EPERM;
3825 } else if (rule->pending) {
3826 return OFPROTO_POSTPONE;
3828 struct rule_collection rules;
3830 rule_collection_init(&rules);
3831 rule_collection_add(&rules, rule);
3832 fm->modify_cookie = true;
3833 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3834 rule_collection_destroy(&rules);
3840 /* Verify actions. */
3841 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3842 &fm->match.flow, table_id);
3844 cls_rule_destroy(&cr);
3848 /* Serialize against pending deletion. */
3849 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3850 cls_rule_destroy(&cr);
3851 return OFPROTO_POSTPONE;
3854 /* Check for overlap, if requested. */
3855 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3858 ovs_rwlock_rdlock(&table->cls.rwlock);
3859 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3860 ovs_rwlock_unlock(&table->cls.rwlock);
3863 cls_rule_destroy(&cr);
3864 return OFPERR_OFPFMFC_OVERLAP;
3868 /* If necessary, evict an existing rule to clear out space. */
3869 error = evict_rules_from_table(ofproto, table, 1);
3871 cls_rule_destroy(&cr);
3875 /* Allocate new rule. */
3876 rule = ofproto->ofproto_class->rule_alloc();
3878 cls_rule_destroy(&cr);
3879 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3880 ofproto->name, ovs_strerror(error));
3884 /* Initialize base state. */
3885 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
3886 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
3887 atomic_init(&rule->ref_count, 1);
3888 rule->pending = NULL;
3889 rule->flow_cookie = fm->new_cookie;
3890 rule->created = rule->modified = rule->used = time_msec();
3892 ovs_mutex_init(&rule->mutex);
3893 ovs_mutex_lock(&rule->mutex);
3894 rule->idle_timeout = fm->idle_timeout;
3895 rule->hard_timeout = fm->hard_timeout;
3896 ovs_mutex_unlock(&rule->mutex);
3898 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
3899 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3900 rule->actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
3901 list_init(&rule->meter_list_node);
3902 rule->eviction_group = NULL;
3903 list_init(&rule->expirable);
3904 rule->monitor_flags = 0;
3905 rule->add_seqno = 0;
3906 rule->modify_seqno = 0;
3908 /* Construct rule, initializing derived state. */
3909 error = ofproto->ofproto_class->rule_construct(rule);
3911 ofproto_rule_destroy__(rule);
3916 oftable_insert_rule(rule);
3918 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3919 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3920 ofproto->ofproto_class->rule_insert(rule);
3921 ofopgroup_submit(group);
3926 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3928 /* Modifies the rules listed in 'rules', changing their actions to match those
3931 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3934 * Returns 0 on success, otherwise an OpenFlow error code. */
3936 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3937 struct ofputil_flow_mod *fm, const struct ofp_header *request,
3938 const struct rule_collection *rules)
3939 OVS_REQUIRES(ofproto_mutex)
3941 enum ofoperation_type type;
3942 struct ofopgroup *group;
3946 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
3947 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3948 error = OFPERR_OFPBRC_EPERM;
3949 for (i = 0; i < rules->n; i++) {
3950 struct rule *rule = rules->rules[i];
3951 struct ofoperation *op;
3952 bool actions_changed;
3953 bool reset_counters;
3955 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
3957 if (rule_is_modifiable(rule)) {
3958 /* At least one rule is modifiable, don't report EPERM error. */
3964 /* Verify actions. */
3965 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
3966 u16_to_ofp(ofproto->max_ports), rule->table_id);
3971 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
3972 rule->actions->ofpacts,
3973 rule->actions->ofpacts_len);
3975 op = ofoperation_create(group, rule, type, 0);
3977 if (fm->modify_cookie && fm->new_cookie != htonll(UINT64_MAX)) {
3978 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
3980 if (type == OFOPERATION_REPLACE) {
3981 ovs_mutex_lock(&rule->mutex);
3982 rule->idle_timeout = fm->idle_timeout;
3983 rule->hard_timeout = fm->hard_timeout;
3984 ovs_mutex_unlock(&rule->mutex);
3986 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3987 if (fm->idle_timeout || fm->hard_timeout) {
3988 if (!rule->eviction_group) {
3989 eviction_group_add_rule(rule);
3992 eviction_group_remove_rule(rule);
3996 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
3997 if (actions_changed || reset_counters) {
3998 struct rule_actions *new_actions;
4000 op->actions = rule->actions;
4001 new_actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4003 ovs_mutex_lock(&rule->mutex);
4004 rule->actions = new_actions;
4005 ovs_mutex_unlock(&rule->mutex);
4007 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4010 ofoperation_complete(op, 0);
4013 ofopgroup_submit(group);
4019 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4020 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4021 OVS_REQUIRES(ofproto_mutex)
4023 if (fm->cookie_mask != htonll(0) || fm->new_cookie == htonll(UINT64_MAX)) {
4026 return add_flow(ofproto, ofconn, fm, request);
4029 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4032 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4035 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4036 struct ofputil_flow_mod *fm,
4037 const struct ofp_header *request)
4038 OVS_REQUIRES(ofproto_mutex)
4040 struct rule_criteria criteria;
4041 struct rule_collection rules;
4044 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4045 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4046 error = collect_rules_loose(ofproto, &criteria, &rules);
4047 rule_criteria_destroy(&criteria);
4050 error = (rules.n > 0
4051 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4052 : modify_flows_add(ofproto, ofconn, fm, request));
4055 rule_collection_destroy(&rules);
4060 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4063 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4066 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4067 struct ofputil_flow_mod *fm,
4068 const struct ofp_header *request)
4069 OVS_REQUIRES(ofproto_mutex)
4071 struct rule_criteria criteria;
4072 struct rule_collection rules;
4075 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4076 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4077 error = collect_rules_strict(ofproto, &criteria, &rules);
4078 rule_criteria_destroy(&criteria);
4082 error = modify_flows_add(ofproto, ofconn, fm, request);
4083 } else if (rules.n == 1) {
4084 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4088 rule_collection_destroy(&rules);
4093 /* OFPFC_DELETE implementation. */
4096 delete_flow__(struct rule *rule, struct ofopgroup *group,
4097 enum ofp_flow_removed_reason reason)
4098 OVS_REQUIRES(ofproto_mutex)
4100 struct ofproto *ofproto = rule->ofproto;
4102 ofproto_rule_send_removed(rule, reason);
4104 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4105 oftable_remove_rule(rule);
4106 ofproto->ofproto_class->rule_delete(rule);
4109 /* Deletes the rules listed in 'rules'.
4111 * Returns 0 on success, otherwise an OpenFlow error code. */
4113 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4114 const struct ofp_header *request,
4115 const struct rule_collection *rules,
4116 enum ofp_flow_removed_reason reason)
4117 OVS_REQUIRES(ofproto_mutex)
4119 struct ofopgroup *group;
4122 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4123 for (i = 0; i < rules->n; i++) {
4124 delete_flow__(rules->rules[i], group, reason);
4126 ofopgroup_submit(group);
4131 /* Implements OFPFC_DELETE. */
4133 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4134 const struct ofputil_flow_mod *fm,
4135 const struct ofp_header *request)
4136 OVS_REQUIRES(ofproto_mutex)
4138 struct rule_criteria criteria;
4139 struct rule_collection rules;
4142 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4143 fm->cookie, fm->cookie_mask,
4144 fm->out_port, fm->out_group);
4145 error = collect_rules_loose(ofproto, &criteria, &rules);
4146 rule_criteria_destroy(&criteria);
4148 if (!error && rules.n > 0) {
4149 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4151 rule_collection_destroy(&rules);
4156 /* Implements OFPFC_DELETE_STRICT. */
4158 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4159 const struct ofputil_flow_mod *fm,
4160 const struct ofp_header *request)
4161 OVS_REQUIRES(ofproto_mutex)
4163 struct rule_criteria criteria;
4164 struct rule_collection rules;
4167 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4168 fm->cookie, fm->cookie_mask,
4169 fm->out_port, fm->out_group);
4170 error = collect_rules_strict(ofproto, &criteria, &rules);
4171 rule_criteria_destroy(&criteria);
4173 if (!error && rules.n > 0) {
4174 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4176 rule_collection_destroy(&rules);
4182 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4183 OVS_REQUIRES(ofproto_mutex)
4185 struct ofputil_flow_removed fr;
4187 if (ofproto_rule_is_hidden(rule) ||
4188 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4192 minimatch_expand(&rule->cr.match, &fr.match);
4193 fr.priority = rule->cr.priority;
4194 fr.cookie = rule->flow_cookie;
4196 fr.table_id = rule->table_id;
4197 calc_duration(rule->created, time_msec(),
4198 &fr.duration_sec, &fr.duration_nsec);
4199 ovs_mutex_lock(&rule->mutex);
4200 fr.idle_timeout = rule->idle_timeout;
4201 fr.hard_timeout = rule->hard_timeout;
4202 ovs_mutex_unlock(&rule->mutex);
4203 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4206 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4209 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4210 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4213 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4216 * ofproto implementation ->run() functions should use this function to expire
4217 * OpenFlow flows. */
4219 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4220 OVS_REQUIRES(ofproto_mutex)
4222 struct ofproto *ofproto = rule->ofproto;
4224 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4225 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4227 ofproto_rule_send_removed(rule, reason);
4228 ofproto_rule_delete__(ofproto, rule);
4231 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4232 * means "infinite". */
4234 reduce_timeout(uint16_t max, uint16_t *timeout)
4236 if (max && (!*timeout || *timeout > max)) {
4241 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4242 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4243 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4245 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4247 ofproto_rule_reduce_timeouts(struct rule *rule,
4248 uint16_t idle_timeout, uint16_t hard_timeout)
4249 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4251 if (!idle_timeout && !hard_timeout) {
4255 ovs_mutex_lock(&ofproto_mutex);
4256 if (list_is_empty(&rule->expirable)) {
4257 list_insert(&rule->ofproto->expirable, &rule->expirable);
4259 ovs_mutex_unlock(&ofproto_mutex);
4261 ovs_mutex_lock(&rule->mutex);
4262 reduce_timeout(idle_timeout, &rule->idle_timeout);
4263 reduce_timeout(hard_timeout, &rule->hard_timeout);
4264 ovs_mutex_unlock(&rule->mutex);
4268 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4269 OVS_EXCLUDED(ofproto_mutex)
4271 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4272 struct ofputil_flow_mod fm;
4273 uint64_t ofpacts_stub[1024 / 8];
4274 struct ofpbuf ofpacts;
4278 error = reject_slave_controller(ofconn);
4283 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4284 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4287 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4290 goto exit_free_ofpacts;
4293 /* Record the operation for logging a summary report. */
4294 switch (fm.command) {
4300 case OFPFC_MODIFY_STRICT:
4301 ofproto->n_modify++;
4305 case OFPFC_DELETE_STRICT:
4306 ofproto->n_delete++;
4311 if (ofproto->next_op_report == LLONG_MAX) {
4312 ofproto->first_op = now;
4313 ofproto->next_op_report = MAX(now + 10 * 1000,
4314 ofproto->op_backoff);
4315 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4317 ofproto->last_op = now;
4320 ofpbuf_uninit(&ofpacts);
4326 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4327 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4328 OVS_EXCLUDED(ofproto_mutex)
4332 ovs_mutex_lock(&ofproto_mutex);
4333 if (ofproto->n_pending < 50) {
4334 switch (fm->command) {
4336 error = add_flow(ofproto, ofconn, fm, oh);
4340 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4343 case OFPFC_MODIFY_STRICT:
4344 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4348 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4351 case OFPFC_DELETE_STRICT:
4352 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4356 if (fm->command > 0xff) {
4357 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4358 "flow_mod_table_id extension is not enabled",
4361 error = OFPERR_OFPFMFC_BAD_COMMAND;
4365 ovs_assert(!list_is_empty(&ofproto->pending));
4366 error = OFPROTO_POSTPONE;
4368 ovs_mutex_unlock(&ofproto_mutex);
4370 run_rule_executes(ofproto);
4375 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4377 struct ofputil_role_request request;
4378 struct ofputil_role_request reply;
4382 error = ofputil_decode_role_message(oh, &request);
4387 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4388 if (ofconn_get_role(ofconn) != request.role
4389 && ofconn_has_pending_opgroups(ofconn)) {
4390 return OFPROTO_POSTPONE;
4393 if (request.have_generation_id
4394 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4395 return OFPERR_OFPRRFC_STALE;
4398 ofconn_set_role(ofconn, request.role);
4401 reply.role = ofconn_get_role(ofconn);
4402 reply.have_generation_id = ofconn_get_master_election_id(
4403 ofconn, &reply.generation_id);
4404 buf = ofputil_encode_role_reply(oh, &reply);
4405 ofconn_send_reply(ofconn, buf);
4411 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4412 const struct ofp_header *oh)
4414 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4415 enum ofputil_protocol cur, next;
4417 cur = ofconn_get_protocol(ofconn);
4418 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4419 ofconn_set_protocol(ofconn, next);
4425 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4427 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4428 enum ofputil_protocol cur, next;
4429 enum ofputil_protocol next_base;
4431 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4433 return OFPERR_OFPBRC_EPERM;
4436 cur = ofconn_get_protocol(ofconn);
4437 next = ofputil_protocol_set_base(cur, next_base);
4438 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4439 /* Avoid sending async messages in surprising protocol. */
4440 return OFPROTO_POSTPONE;
4443 ofconn_set_protocol(ofconn, next);
4448 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4449 const struct ofp_header *oh)
4451 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4454 format = ntohl(msg->format);
4455 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4456 return OFPERR_OFPBRC_EPERM;
4459 if (format != ofconn_get_packet_in_format(ofconn)
4460 && ofconn_has_pending_opgroups(ofconn)) {
4461 /* Avoid sending async message in surprsing packet in format. */
4462 return OFPROTO_POSTPONE;
4465 ofconn_set_packet_in_format(ofconn, format);
4470 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4472 const struct nx_async_config *msg = ofpmsg_body(oh);
4473 uint32_t master[OAM_N_TYPES];
4474 uint32_t slave[OAM_N_TYPES];
4476 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4477 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4478 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4480 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4481 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4482 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4484 ofconn_set_async_config(ofconn, master, slave);
4485 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4486 !ofconn_get_miss_send_len(ofconn)) {
4487 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4494 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4497 uint32_t master[OAM_N_TYPES];
4498 uint32_t slave[OAM_N_TYPES];
4499 struct nx_async_config *msg;
4501 ofconn_get_async_config(ofconn, master, slave);
4502 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4503 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4505 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4506 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4507 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4509 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4510 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4511 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4513 ofconn_send_reply(ofconn, buf);
4519 handle_nxt_set_controller_id(struct ofconn *ofconn,
4520 const struct ofp_header *oh)
4522 const struct nx_controller_id *nci = ofpmsg_body(oh);
4524 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4525 return OFPERR_NXBRC_MUST_BE_ZERO;
4528 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4533 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4537 if (ofconn_has_pending_opgroups(ofconn)) {
4538 return OFPROTO_POSTPONE;
4541 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4542 ? OFPRAW_OFPT10_BARRIER_REPLY
4543 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4544 ofconn_send_reply(ofconn, buf);
4549 ofproto_compose_flow_refresh_update(const struct rule *rule,
4550 enum nx_flow_monitor_flags flags,
4552 OVS_REQUIRES(ofproto_mutex)
4554 struct ofoperation *op = rule->pending;
4555 const struct rule_actions *actions;
4556 struct ofputil_flow_update fu;
4559 if (op && op->type == OFOPERATION_ADD) {
4560 /* We'll report the final flow when the operation completes. Reporting
4561 * it now would cause a duplicate report later. */
4565 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4566 ? NXFME_ADDED : NXFME_MODIFIED);
4568 ovs_mutex_lock(&rule->mutex);
4569 fu.idle_timeout = rule->idle_timeout;
4570 fu.hard_timeout = rule->hard_timeout;
4571 ovs_mutex_unlock(&rule->mutex);
4572 fu.table_id = rule->table_id;
4573 fu.cookie = rule->flow_cookie;
4574 minimatch_expand(&rule->cr.match, &match);
4576 fu.priority = rule->cr.priority;
4578 if (!(flags & NXFMF_ACTIONS)) {
4581 actions = rule->actions;
4583 /* An operation is in progress. Use the previous version of the flow's
4584 * actions, so that when the operation commits we report the change. */
4586 case OFOPERATION_ADD:
4589 case OFOPERATION_MODIFY:
4590 case OFOPERATION_REPLACE:
4591 actions = op->actions ? op->actions : rule->actions;
4594 case OFOPERATION_DELETE:
4595 actions = rule->actions;
4602 fu.ofpacts = actions ? actions->ofpacts : NULL;
4603 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4605 if (list_is_empty(msgs)) {
4606 ofputil_start_flow_update(msgs);
4608 ofputil_append_flow_update(&fu, msgs);
4612 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4614 OVS_REQUIRES(ofproto_mutex)
4618 for (i = 0; i < rules->n; i++) {
4619 struct rule *rule = rules->rules[i];
4620 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4621 rule->monitor_flags = 0;
4623 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4628 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4629 struct rule *rule, uint64_t seqno,
4630 struct rule_collection *rules)
4631 OVS_REQUIRES(ofproto_mutex)
4633 enum nx_flow_monitor_flags update;
4635 if (ofproto_rule_is_hidden(rule)) {
4640 ? ofoperation_has_out_port(rule->pending, m->out_port)
4641 : ofproto_rule_has_out_port(rule, m->out_port))) {
4646 if (rule->add_seqno > seqno) {
4647 update = NXFMF_ADD | NXFMF_MODIFY;
4648 } else if (rule->modify_seqno > seqno) {
4649 update = NXFMF_MODIFY;
4654 if (!(m->flags & update)) {
4658 update = NXFMF_INITIAL;
4661 if (!rule->monitor_flags) {
4662 rule_collection_add(rules, rule);
4664 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4668 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4670 struct rule_collection *rules)
4671 OVS_REQUIRES(ofproto_mutex)
4673 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4674 const struct ofoperation *op;
4675 const struct oftable *table;
4676 struct cls_rule target;
4678 cls_rule_init_from_minimatch(&target, &m->match, 0);
4679 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4680 struct cls_cursor cursor;
4683 ovs_rwlock_rdlock(&table->cls.rwlock);
4684 cls_cursor_init(&cursor, &table->cls, &target);
4685 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4686 ovs_assert(!rule->pending); /* XXX */
4687 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4689 ovs_rwlock_unlock(&table->cls.rwlock);
4692 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4693 struct rule *rule = op->rule;
4695 if (((m->table_id == 0xff
4696 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4697 : m->table_id == rule->table_id))
4698 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4699 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4702 cls_rule_destroy(&target);
4706 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4707 struct rule_collection *rules)
4708 OVS_REQUIRES(ofproto_mutex)
4710 if (m->flags & NXFMF_INITIAL) {
4711 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4716 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4717 uint64_t seqno, struct rule_collection *rules)
4718 OVS_REQUIRES(ofproto_mutex)
4720 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4724 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4725 OVS_EXCLUDED(ofproto_mutex)
4727 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4728 struct ofmonitor **monitors;
4729 size_t n_monitors, allocated_monitors;
4730 struct rule_collection rules;
4731 struct list replies;
4737 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4739 n_monitors = allocated_monitors = 0;
4741 ovs_mutex_lock(&ofproto_mutex);
4743 struct ofputil_flow_monitor_request request;
4744 struct ofmonitor *m;
4747 retval = ofputil_decode_flow_monitor_request(&request, &b);
4748 if (retval == EOF) {
4750 } else if (retval) {
4755 if (request.table_id != 0xff
4756 && request.table_id >= ofproto->n_tables) {
4757 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4761 error = ofmonitor_create(&request, ofconn, &m);
4766 if (n_monitors >= allocated_monitors) {
4767 monitors = x2nrealloc(monitors, &allocated_monitors,
4770 monitors[n_monitors++] = m;
4773 rule_collection_init(&rules);
4774 for (i = 0; i < n_monitors; i++) {
4775 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4778 ofpmp_init(&replies, oh);
4779 ofmonitor_compose_refresh_updates(&rules, &replies);
4780 ovs_mutex_unlock(&ofproto_mutex);
4782 rule_collection_destroy(&rules);
4784 ofconn_send_replies(ofconn, &replies);
4790 ovs_mutex_unlock(&ofproto_mutex);
4792 for (i = 0; i < n_monitors; i++) {
4793 ofmonitor_destroy(monitors[i]);
4800 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4801 OVS_EXCLUDED(ofproto_mutex)
4803 struct ofmonitor *m;
4807 id = ofputil_decode_flow_monitor_cancel(oh);
4809 ovs_mutex_lock(&ofproto_mutex);
4810 m = ofmonitor_lookup(ofconn, id);
4812 ofmonitor_destroy(m);
4815 error = OFPERR_NXBRC_FM_BAD_ID;
4817 ovs_mutex_unlock(&ofproto_mutex);
4822 /* Meters implementation.
4824 * Meter table entry, indexed by the OpenFlow meter_id.
4825 * These are always dynamically allocated to allocate enough space for
4827 * 'created' is used to compute the duration for meter stats.
4828 * 'list rules' is needed so that we can delete the dependent rules when the
4829 * meter table entry is deleted.
4830 * 'provider_meter_id' is for the provider's private use.
4833 long long int created; /* Time created. */
4834 struct list rules; /* List of "struct rule_dpif"s. */
4835 ofproto_meter_id provider_meter_id;
4836 uint16_t flags; /* Meter flags. */
4837 uint16_t n_bands; /* Number of meter bands. */
4838 struct ofputil_meter_band *bands;
4842 * This is used in instruction validation at flow set-up time,
4843 * as flows may not use non-existing meters.
4844 * This is also used by ofproto-providers to translate OpenFlow meter_ids
4845 * in METER instructions to the corresponding provider meter IDs.
4846 * Return value of UINT32_MAX signifies an invalid meter.
4849 ofproto_get_provider_meter_id(const struct ofproto * ofproto,
4850 uint32_t of_meter_id)
4852 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4853 const struct meter *meter = ofproto->meters[of_meter_id];
4855 return meter->provider_meter_id.uint32;
4862 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4866 meter->flags = config->flags;
4867 meter->n_bands = config->n_bands;
4868 meter->bands = xmemdup(config->bands,
4869 config->n_bands * sizeof *meter->bands);
4872 static struct meter *
4873 meter_create(const struct ofputil_meter_config *config,
4874 ofproto_meter_id provider_meter_id)
4876 struct meter *meter;
4878 meter = xzalloc(sizeof *meter);
4879 meter->provider_meter_id = provider_meter_id;
4880 meter->created = time_msec();
4881 list_init(&meter->rules);
4883 meter_update(meter, config);
4889 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4890 OVS_REQUIRES(ofproto_mutex)
4893 for (mid = first; mid <= last; ++mid) {
4894 struct meter *meter = ofproto->meters[mid];
4896 ofproto->meters[mid] = NULL;
4897 ofproto->ofproto_class->meter_del(ofproto,
4898 meter->provider_meter_id);
4906 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4908 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4909 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4913 return OFPERR_OFPMMFC_METER_EXISTS;
4916 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4919 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4920 *meterp = meter_create(&mm->meter, provider_meter_id);
4926 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4928 struct meter *meter = ofproto->meters[mm->meter.meter_id];
4932 return OFPERR_OFPMMFC_UNKNOWN_METER;
4935 error = ofproto->ofproto_class->meter_set(ofproto,
4936 &meter->provider_meter_id,
4938 ovs_assert(meter->provider_meter_id.uint32 != UINT32_MAX);
4940 meter_update(meter, &mm->meter);
4946 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
4947 struct ofputil_meter_mod *mm)
4948 OVS_EXCLUDED(ofproto_mutex)
4950 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4951 uint32_t meter_id = mm->meter.meter_id;
4952 struct rule_collection rules;
4953 enum ofperr error = 0;
4954 uint32_t first, last;
4956 if (meter_id == OFPM13_ALL) {
4958 last = ofproto->meter_features.max_meters;
4960 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
4963 first = last = meter_id;
4966 /* First delete the rules that use this meter. If any of those rules are
4967 * currently being modified, postpone the whole operation until later. */
4968 rule_collection_init(&rules);
4969 ovs_mutex_lock(&ofproto_mutex);
4970 for (meter_id = first; meter_id <= last; ++meter_id) {
4971 struct meter *meter = ofproto->meters[meter_id];
4972 if (meter && !list_is_empty(&meter->rules)) {
4975 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
4976 if (rule->pending) {
4977 error = OFPROTO_POSTPONE;
4980 rule_collection_add(&rules, rule);
4985 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
4988 /* Delete the meters. */
4989 meter_delete(ofproto, first, last);
4992 ovs_mutex_unlock(&ofproto_mutex);
4993 rule_collection_destroy(&rules);
4999 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5001 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5002 struct ofputil_meter_mod mm;
5003 uint64_t bands_stub[256 / 8];
5004 struct ofpbuf bands;
5008 error = reject_slave_controller(ofconn);
5013 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5015 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5017 goto exit_free_bands;
5020 meter_id = mm.meter.meter_id;
5022 if (mm.command != OFPMC13_DELETE) {
5023 /* Fails also when meters are not implemented by the provider. */
5024 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5025 error = OFPERR_OFPMMFC_INVALID_METER;
5026 goto exit_free_bands;
5027 } else if (meter_id > ofproto->meter_features.max_meters) {
5028 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5029 goto exit_free_bands;
5031 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5032 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5033 goto exit_free_bands;
5037 switch (mm.command) {
5039 error = handle_add_meter(ofproto, &mm);
5042 case OFPMC13_MODIFY:
5043 error = handle_modify_meter(ofproto, &mm);
5046 case OFPMC13_DELETE:
5047 error = handle_delete_meter(ofconn, oh, &mm);
5051 error = OFPERR_OFPMMFC_BAD_COMMAND;
5056 ofpbuf_uninit(&bands);
5061 handle_meter_features_request(struct ofconn *ofconn,
5062 const struct ofp_header *request)
5064 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5065 struct ofputil_meter_features features;
5068 if (ofproto->ofproto_class->meter_get_features) {
5069 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5071 memset(&features, 0, sizeof features);
5073 b = ofputil_encode_meter_features_reply(&features, request);
5075 ofconn_send_reply(ofconn, b);
5080 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5083 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5084 struct list replies;
5085 uint64_t bands_stub[256 / 8];
5086 struct ofpbuf bands;
5087 uint32_t meter_id, first, last;
5089 ofputil_decode_meter_request(request, &meter_id);
5091 if (meter_id == OFPM13_ALL) {
5093 last = ofproto->meter_features.max_meters;
5095 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5096 !ofproto->meters[meter_id]) {
5097 return OFPERR_OFPMMFC_UNKNOWN_METER;
5099 first = last = meter_id;
5102 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5103 ofpmp_init(&replies, request);
5105 for (meter_id = first; meter_id <= last; ++meter_id) {
5106 struct meter *meter = ofproto->meters[meter_id];
5108 continue; /* Skip non-existing meters. */
5110 if (type == OFPTYPE_METER_STATS_REQUEST) {
5111 struct ofputil_meter_stats stats;
5113 stats.meter_id = meter_id;
5115 /* Provider sets the packet and byte counts, we do the rest. */
5116 stats.flow_count = list_size(&meter->rules);
5117 calc_duration(meter->created, time_msec(),
5118 &stats.duration_sec, &stats.duration_nsec);
5119 stats.n_bands = meter->n_bands;
5120 ofpbuf_clear(&bands);
5122 = ofpbuf_put_uninit(&bands,
5123 meter->n_bands * sizeof *stats.bands);
5125 if (!ofproto->ofproto_class->meter_get(ofproto,
5126 meter->provider_meter_id,
5128 ofputil_append_meter_stats(&replies, &stats);
5130 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5131 struct ofputil_meter_config config;
5133 config.meter_id = meter_id;
5134 config.flags = meter->flags;
5135 config.n_bands = meter->n_bands;
5136 config.bands = meter->bands;
5137 ofputil_append_meter_config(&replies, &config);
5141 ofconn_send_replies(ofconn, &replies);
5142 ofpbuf_uninit(&bands);
5147 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5148 struct ofgroup **group)
5149 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5151 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5152 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5153 hash_int(group_id, 0), &ofproto->groups) {
5154 if ((*group)->group_id == group_id) {
5155 ovs_rwlock_rdlock(&(*group)->rwlock);
5156 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5160 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5165 ofproto_group_release(struct ofgroup *group)
5166 OVS_RELEASES(group->rwlock)
5168 ovs_rwlock_unlock(&group->rwlock);
5172 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5173 struct ofgroup **group)
5174 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5175 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5177 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5178 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5179 hash_int(group_id, 0), &ofproto->groups) {
5180 if ((*group)->group_id == group_id) {
5181 ovs_rwlock_wrlock(&(*group)->rwlock);
5185 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5190 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5191 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5193 struct ofgroup *grp;
5195 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5196 hash_int(group_id, 0), &ofproto->groups) {
5197 if (grp->group_id == group_id) {
5205 append_group_stats(struct ofgroup *group, struct list *replies)
5206 OVS_REQ_RDLOCK(group->rwlock)
5208 struct ofputil_group_stats ogs;
5209 struct ofproto *ofproto = group->ofproto;
5210 long long int now = time_msec();
5213 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5215 error = (ofproto->ofproto_class->group_get_stats
5216 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5219 ogs.ref_count = UINT32_MAX;
5220 ogs.packet_count = UINT64_MAX;
5221 ogs.byte_count = UINT64_MAX;
5222 ogs.n_buckets = group->n_buckets;
5223 memset(ogs.bucket_stats, 0xff,
5224 ogs.n_buckets * sizeof *ogs.bucket_stats);
5227 ogs.group_id = group->group_id;
5228 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5230 ofputil_append_group_stats(replies, &ogs);
5232 free(ogs.bucket_stats);
5236 handle_group_stats_request(struct ofconn *ofconn,
5237 const struct ofp_header *request)
5239 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5240 struct list replies;
5242 struct ofgroup *group;
5245 error = ofputil_decode_group_stats_request(request, &group_id);
5250 ofpmp_init(&replies, request);
5252 if (group_id == OFPG_ALL) {
5253 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5254 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5255 ovs_rwlock_rdlock(&group->rwlock);
5256 append_group_stats(group, &replies);
5257 ovs_rwlock_unlock(&group->rwlock);
5259 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5261 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5262 append_group_stats(group, &replies);
5263 ofproto_group_release(group);
5267 ofconn_send_replies(ofconn, &replies);
5273 handle_group_desc_stats_request(struct ofconn *ofconn,
5274 const struct ofp_header *request)
5276 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5277 struct list replies;
5278 struct ofputil_group_desc gds;
5279 struct ofgroup *group;
5281 ofpmp_init(&replies, request);
5283 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5284 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5285 gds.group_id = group->group_id;
5286 gds.type = group->type;
5287 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5289 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5291 ofconn_send_replies(ofconn, &replies);
5297 handle_group_features_stats_request(struct ofconn *ofconn,
5298 const struct ofp_header *request)
5300 struct ofproto *p = ofconn_get_ofproto(ofconn);
5303 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5305 ofconn_send_reply(ofconn, msg);
5311 /* Implements OFPGC11_ADD
5312 * in which no matching flow already exists in the flow table.
5314 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5315 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5316 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5317 * initiated now but may be retried later.
5319 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5320 * ownership remains with the caller.
5322 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5325 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5327 struct ofgroup *ofgroup;
5330 if (gm->group_id > OFPG_MAX) {
5331 return OFPERR_OFPGMFC_INVALID_GROUP;
5333 if (gm->type > OFPGT11_FF) {
5334 return OFPERR_OFPGMFC_BAD_TYPE;
5337 /* Allocate new group and initialize it. */
5338 ofgroup = ofproto->ofproto_class->group_alloc();
5340 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5341 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5344 ovs_rwlock_init(&ofgroup->rwlock);
5345 ofgroup->ofproto = ofproto;
5346 ofgroup->group_id = gm->group_id;
5347 ofgroup->type = gm->type;
5348 ofgroup->created = ofgroup->modified = time_msec();
5350 list_move(&ofgroup->buckets, &gm->buckets);
5351 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5353 /* Construct called BEFORE any locks are held. */
5354 error = ofproto->ofproto_class->group_construct(ofgroup);
5359 /* We wrlock as late as possible to minimize the time we jam any other
5360 * threads: No visible state changes before acquiring the lock. */
5361 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5363 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5364 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5368 if (ofproto_group_exists(ofproto, gm->group_id)) {
5369 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5374 /* Insert new group. */
5375 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5376 hash_int(ofgroup->group_id, 0));
5377 ofproto->n_groups[ofgroup->type]++;
5379 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5384 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5385 ofproto->ofproto_class->group_destruct(ofgroup);
5387 ofputil_bucket_list_destroy(&ofgroup->buckets);
5388 ofproto->ofproto_class->group_dealloc(ofgroup);
5393 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5396 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5399 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5401 struct ofgroup *ofgroup;
5402 struct ofgroup *victim;
5405 if (gm->group_id > OFPG_MAX) {
5406 return OFPERR_OFPGMFC_INVALID_GROUP;
5409 if (gm->type > OFPGT11_FF) {
5410 return OFPERR_OFPGMFC_BAD_TYPE;
5413 victim = ofproto->ofproto_class->group_alloc();
5415 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5416 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5419 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5420 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5423 /* Both group's and its container's write locks held now.
5424 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5425 if (ofgroup->type != gm->type
5426 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5427 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5432 list_move(&victim->buckets, &ofgroup->buckets);
5434 ofgroup->type = gm->type;
5435 list_move(&ofgroup->buckets, &gm->buckets);
5436 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5438 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5440 ofputil_bucket_list_destroy(&victim->buckets);
5441 ofproto->n_groups[victim->type]--;
5442 ofproto->n_groups[ofgroup->type]++;
5443 ofgroup->modified = time_msec();
5445 ofputil_bucket_list_destroy(&ofgroup->buckets);
5448 list_move(&ofgroup->buckets, &victim->buckets);
5452 ovs_rwlock_unlock(&ofgroup->rwlock);
5453 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5455 ofproto->ofproto_class->group_dealloc(victim);
5460 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5461 OVS_RELEASES(ofproto->groups_rwlock)
5463 /* Must wait until existing readers are done,
5464 * while holding the container's write lock at the same time. */
5465 ovs_rwlock_wrlock(&ofgroup->rwlock);
5466 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5467 /* No-one can find this group any more. */
5468 ofproto->n_groups[ofgroup->type]--;
5469 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5471 ofproto->ofproto_class->group_destruct(ofgroup);
5472 ofputil_bucket_list_destroy(&ofgroup->buckets);
5473 ovs_rwlock_unlock(&ofgroup->rwlock);
5474 ovs_rwlock_destroy(&ofgroup->rwlock);
5475 ofproto->ofproto_class->group_dealloc(ofgroup);
5478 /* Implements OFPGC_DELETE. */
5480 delete_group(struct ofproto *ofproto, uint32_t group_id)
5482 struct ofgroup *ofgroup;
5484 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5485 if (group_id == OFPG_ALL) {
5487 struct hmap_node *node = hmap_first(&ofproto->groups);
5491 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5492 delete_group__(ofproto, ofgroup);
5493 /* Lock for each node separately, so that we will not jam the
5494 * other threads for too long time. */
5495 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5498 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5499 hash_int(group_id, 0), &ofproto->groups) {
5500 if (ofgroup->group_id == group_id) {
5501 delete_group__(ofproto, ofgroup);
5506 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5510 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5512 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5513 struct ofputil_group_mod gm;
5516 error = reject_slave_controller(ofconn);
5521 error = ofputil_decode_group_mod(oh, &gm);
5526 switch (gm.command) {
5528 return add_group(ofproto, &gm);
5530 case OFPGC11_MODIFY:
5531 return modify_group(ofproto, &gm);
5533 case OFPGC11_DELETE:
5534 delete_group(ofproto, gm.group_id);
5538 if (gm.command > OFPGC11_DELETE) {
5539 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5540 ofproto->name, gm.command);
5542 return OFPERR_OFPGMFC_BAD_COMMAND;
5547 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5549 struct ofputil_table_mod tm;
5552 error = reject_slave_controller(ofconn);
5557 error = ofputil_decode_table_mod(oh, &tm);
5562 /* XXX Actual table mod support is not implemented yet. */
5567 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5568 OVS_EXCLUDED(ofproto_mutex)
5570 const struct ofp_header *oh = msg->data;
5574 error = ofptype_decode(&type, oh);
5580 /* OpenFlow requests. */
5581 case OFPTYPE_ECHO_REQUEST:
5582 return handle_echo_request(ofconn, oh);
5584 case OFPTYPE_FEATURES_REQUEST:
5585 return handle_features_request(ofconn, oh);
5587 case OFPTYPE_GET_CONFIG_REQUEST:
5588 return handle_get_config_request(ofconn, oh);
5590 case OFPTYPE_SET_CONFIG:
5591 return handle_set_config(ofconn, oh);
5593 case OFPTYPE_PACKET_OUT:
5594 return handle_packet_out(ofconn, oh);
5596 case OFPTYPE_PORT_MOD:
5597 return handle_port_mod(ofconn, oh);
5599 case OFPTYPE_FLOW_MOD:
5600 return handle_flow_mod(ofconn, oh);
5602 case OFPTYPE_GROUP_MOD:
5603 return handle_group_mod(ofconn, oh);
5605 case OFPTYPE_TABLE_MOD:
5606 return handle_table_mod(ofconn, oh);
5608 case OFPTYPE_METER_MOD:
5609 return handle_meter_mod(ofconn, oh);
5611 case OFPTYPE_BARRIER_REQUEST:
5612 return handle_barrier_request(ofconn, oh);
5614 case OFPTYPE_ROLE_REQUEST:
5615 return handle_role_request(ofconn, oh);
5617 /* OpenFlow replies. */
5618 case OFPTYPE_ECHO_REPLY:
5621 /* Nicira extension requests. */
5622 case OFPTYPE_FLOW_MOD_TABLE_ID:
5623 return handle_nxt_flow_mod_table_id(ofconn, oh);
5625 case OFPTYPE_SET_FLOW_FORMAT:
5626 return handle_nxt_set_flow_format(ofconn, oh);
5628 case OFPTYPE_SET_PACKET_IN_FORMAT:
5629 return handle_nxt_set_packet_in_format(ofconn, oh);
5631 case OFPTYPE_SET_CONTROLLER_ID:
5632 return handle_nxt_set_controller_id(ofconn, oh);
5634 case OFPTYPE_FLOW_AGE:
5635 /* Nothing to do. */
5638 case OFPTYPE_FLOW_MONITOR_CANCEL:
5639 return handle_flow_monitor_cancel(ofconn, oh);
5641 case OFPTYPE_SET_ASYNC_CONFIG:
5642 return handle_nxt_set_async_config(ofconn, oh);
5644 case OFPTYPE_GET_ASYNC_REQUEST:
5645 return handle_nxt_get_async_request(ofconn, oh);
5647 /* Statistics requests. */
5648 case OFPTYPE_DESC_STATS_REQUEST:
5649 return handle_desc_stats_request(ofconn, oh);
5651 case OFPTYPE_FLOW_STATS_REQUEST:
5652 return handle_flow_stats_request(ofconn, oh);
5654 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5655 return handle_aggregate_stats_request(ofconn, oh);
5657 case OFPTYPE_TABLE_STATS_REQUEST:
5658 return handle_table_stats_request(ofconn, oh);
5660 case OFPTYPE_PORT_STATS_REQUEST:
5661 return handle_port_stats_request(ofconn, oh);
5663 case OFPTYPE_QUEUE_STATS_REQUEST:
5664 return handle_queue_stats_request(ofconn, oh);
5666 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5667 return handle_port_desc_stats_request(ofconn, oh);
5669 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5670 return handle_flow_monitor_request(ofconn, oh);
5672 case OFPTYPE_METER_STATS_REQUEST:
5673 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5674 return handle_meter_request(ofconn, oh, type);
5676 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5677 return handle_meter_features_request(ofconn, oh);
5679 case OFPTYPE_GROUP_STATS_REQUEST:
5680 return handle_group_stats_request(ofconn, oh);
5682 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5683 return handle_group_desc_stats_request(ofconn, oh);
5685 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5686 return handle_group_features_stats_request(ofconn, oh);
5688 /* FIXME: Change the following once they are implemented: */
5689 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5690 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5691 return OFPERR_OFPBRC_BAD_TYPE;
5695 case OFPTYPE_FEATURES_REPLY:
5696 case OFPTYPE_GET_CONFIG_REPLY:
5697 case OFPTYPE_PACKET_IN:
5698 case OFPTYPE_FLOW_REMOVED:
5699 case OFPTYPE_PORT_STATUS:
5700 case OFPTYPE_BARRIER_REPLY:
5701 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5702 case OFPTYPE_DESC_STATS_REPLY:
5703 case OFPTYPE_FLOW_STATS_REPLY:
5704 case OFPTYPE_QUEUE_STATS_REPLY:
5705 case OFPTYPE_PORT_STATS_REPLY:
5706 case OFPTYPE_TABLE_STATS_REPLY:
5707 case OFPTYPE_AGGREGATE_STATS_REPLY:
5708 case OFPTYPE_PORT_DESC_STATS_REPLY:
5709 case OFPTYPE_ROLE_REPLY:
5710 case OFPTYPE_FLOW_MONITOR_PAUSED:
5711 case OFPTYPE_FLOW_MONITOR_RESUMED:
5712 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5713 case OFPTYPE_GET_ASYNC_REPLY:
5714 case OFPTYPE_GROUP_STATS_REPLY:
5715 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5716 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5717 case OFPTYPE_METER_STATS_REPLY:
5718 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5719 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5720 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5722 return OFPERR_OFPBRC_BAD_TYPE;
5727 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5728 OVS_EXCLUDED(ofproto_mutex)
5730 int error = handle_openflow__(ofconn, ofp_msg);
5731 if (error && error != OFPROTO_POSTPONE) {
5732 ofconn_send_error(ofconn, ofp_msg->data, error);
5734 COVERAGE_INC(ofproto_recv_openflow);
5735 return error != OFPROTO_POSTPONE;
5738 /* Asynchronous operations. */
5740 /* Creates and returns a new ofopgroup that is not associated with any
5741 * OpenFlow connection.
5743 * The caller should add operations to the returned group with
5744 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5745 static struct ofopgroup *
5746 ofopgroup_create_unattached(struct ofproto *ofproto)
5747 OVS_REQUIRES(ofproto_mutex)
5749 struct ofopgroup *group = xzalloc(sizeof *group);
5750 group->ofproto = ofproto;
5751 list_init(&group->ofproto_node);
5752 list_init(&group->ops);
5753 list_init(&group->ofconn_node);
5757 /* Creates and returns a new ofopgroup for 'ofproto'.
5759 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5760 * connection. The 'request' and 'buffer_id' arguments are ignored.
5762 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5763 * If the ofopgroup eventually fails, then the error reply will include
5764 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5765 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5767 * The caller should add operations to the returned group with
5768 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5769 static struct ofopgroup *
5770 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5771 const struct ofp_header *request, uint32_t buffer_id)
5772 OVS_REQUIRES(ofproto_mutex)
5774 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5776 size_t request_len = ntohs(request->length);
5778 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5780 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5781 group->ofconn = ofconn;
5782 group->request = xmemdup(request, MIN(request_len, 64));
5783 group->buffer_id = buffer_id;
5788 /* Submits 'group' for processing.
5790 * If 'group' contains no operations (e.g. none were ever added, or all of the
5791 * ones that were added completed synchronously), then it is destroyed
5792 * immediately. Otherwise it is added to the ofproto's list of pending
5795 ofopgroup_submit(struct ofopgroup *group)
5796 OVS_REQUIRES(ofproto_mutex)
5798 if (!group->n_running) {
5799 ofopgroup_complete(group);
5801 list_push_back(&group->ofproto->pending, &group->ofproto_node);
5802 group->ofproto->n_pending++;
5807 ofopgroup_complete(struct ofopgroup *group)
5808 OVS_REQUIRES(ofproto_mutex)
5810 struct ofproto *ofproto = group->ofproto;
5812 struct ofconn *abbrev_ofconn;
5813 ovs_be32 abbrev_xid;
5815 struct ofoperation *op, *next_op;
5818 ovs_assert(!group->n_running);
5821 LIST_FOR_EACH (op, group_node, &group->ops) {
5828 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
5829 LIST_FOR_EACH (op, group_node, &group->ops) {
5830 if (op->type != OFOPERATION_DELETE) {
5831 struct ofpbuf *packet;
5834 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
5837 struct rule_execute *re;
5841 ofproto_rule_ref(op->rule);
5843 re = xmalloc(sizeof *re);
5844 re->rule = op->rule;
5845 re->in_port = in_port;
5846 re->packet = packet;
5848 if (!guarded_list_push_back(&ofproto->rule_executes,
5849 &re->list_node, 1024)) {
5850 ofproto_rule_unref(op->rule);
5851 ofpbuf_delete(re->packet);
5860 if (!error && !list_is_empty(&group->ofconn_node)) {
5861 abbrev_ofconn = group->ofconn;
5862 abbrev_xid = group->request->xid;
5864 abbrev_ofconn = NULL;
5865 abbrev_xid = htonl(0);
5867 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
5868 struct rule *rule = op->rule;
5870 /* We generally want to report the change to active OpenFlow flow
5871 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5873 - The operation failed.
5875 - The affected rule is not visible to controllers.
5877 - The operation's only effect was to update rule->modified. */
5879 || ofproto_rule_is_hidden(rule)
5880 || (op->type == OFOPERATION_MODIFY
5882 && rule->flow_cookie == op->flow_cookie))) {
5883 /* Check that we can just cast from ofoperation_type to
5884 * nx_flow_update_event. */
5885 enum nx_flow_update_event event_type;
5888 case OFOPERATION_ADD:
5889 case OFOPERATION_REPLACE:
5890 event_type = NXFME_ADDED;
5893 case OFOPERATION_DELETE:
5894 event_type = NXFME_DELETED;
5897 case OFOPERATION_MODIFY:
5898 event_type = NXFME_MODIFIED;
5905 ofmonitor_report(ofproto->connmgr, rule, event_type,
5906 op->reason, abbrev_ofconn, abbrev_xid);
5909 rule->pending = NULL;
5912 case OFOPERATION_ADD:
5916 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5917 if (vid_mask == VLAN_VID_MASK) {
5918 if (ofproto->vlan_bitmap) {
5919 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5920 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5921 bitmap_set1(ofproto->vlan_bitmap, vid);
5922 ofproto->vlans_changed = true;
5925 ofproto->vlans_changed = true;
5929 oftable_remove_rule(rule);
5930 ofproto_rule_unref(rule);
5934 case OFOPERATION_DELETE:
5935 ovs_assert(!op->error);
5936 ofproto_rule_unref(rule);
5940 case OFOPERATION_MODIFY:
5941 case OFOPERATION_REPLACE:
5943 long long int now = time_msec();
5945 rule->modified = now;
5946 if (op->type == OFOPERATION_REPLACE) {
5947 rule->created = rule->used = now;
5950 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
5951 ovs_mutex_lock(&rule->mutex);
5952 rule->idle_timeout = op->idle_timeout;
5953 rule->hard_timeout = op->hard_timeout;
5954 ovs_mutex_unlock(&rule->mutex);
5956 struct rule_actions *old_actions;
5958 ovs_mutex_lock(&rule->mutex);
5959 old_actions = rule->actions;
5960 rule->actions = op->actions;
5961 ovs_mutex_unlock(&rule->mutex);
5964 rule_actions_unref(old_actions);
5966 rule->flags = op->flags;
5974 ofoperation_destroy(op);
5977 ofmonitor_flush(ofproto->connmgr);
5979 if (!list_is_empty(&group->ofproto_node)) {
5980 ovs_assert(ofproto->n_pending > 0);
5981 ofproto->n_pending--;
5982 list_remove(&group->ofproto_node);
5984 if (!list_is_empty(&group->ofconn_node)) {
5985 list_remove(&group->ofconn_node);
5987 ofconn_send_error(group->ofconn, group->request, error);
5989 connmgr_retry(ofproto->connmgr);
5991 free(group->request);
5995 /* Initiates a new operation on 'rule', of the specified 'type', within
5996 * 'group'. Prior to calling, 'rule' must not have any pending operation.
5998 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
5999 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6001 * Returns the newly created ofoperation (which is also available as
6002 * rule->pending). */
6003 static struct ofoperation *
6004 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6005 enum ofoperation_type type,
6006 enum ofp_flow_removed_reason reason)
6007 OVS_REQUIRES(ofproto_mutex)
6009 struct ofproto *ofproto = group->ofproto;
6010 struct ofoperation *op;
6012 ovs_assert(!rule->pending);
6014 op = rule->pending = xzalloc(sizeof *op);
6016 list_push_back(&group->ops, &op->group_node);
6019 op->reason = reason;
6020 op->flow_cookie = rule->flow_cookie;
6021 ovs_mutex_lock(&rule->mutex);
6022 op->idle_timeout = rule->idle_timeout;
6023 op->hard_timeout = rule->hard_timeout;
6024 ovs_mutex_unlock(&rule->mutex);
6025 op->flags = rule->flags;
6029 if (type == OFOPERATION_DELETE) {
6030 hmap_insert(&ofproto->deletions, &op->hmap_node,
6031 cls_rule_hash(&rule->cr, rule->table_id));
6038 ofoperation_destroy(struct ofoperation *op)
6039 OVS_REQUIRES(ofproto_mutex)
6041 struct ofopgroup *group = op->group;
6044 op->rule->pending = NULL;
6046 if (op->type == OFOPERATION_DELETE) {
6047 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6049 list_remove(&op->group_node);
6050 rule_actions_unref(op->actions);
6054 /* Indicates that 'op' completed with status 'error', which is either 0 to
6055 * indicate success or an OpenFlow error code on failure.
6057 * If 'error' is 0, indicating success, the operation will be committed
6058 * permanently to the flow table.
6060 * If 'error' is nonzero, then generally the operation will be rolled back:
6062 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6063 * restores the original rule. The caller must have uninitialized any
6064 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6065 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6066 * and 7 for the new rule, calling its ->rule_dealloc() function.
6068 * - If 'op' is a "modify flow" operation, ofproto restores the original
6071 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6072 * allowed to fail. It must always succeed.
6074 * Please see the large comment in ofproto/ofproto-provider.h titled
6075 * "Asynchronous Operation Support" for more information. */
6077 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6079 struct ofopgroup *group = op->group;
6081 ovs_assert(group->n_running > 0);
6082 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6085 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6086 /* This function can be called from ->rule_construct(), in which case
6087 * ofproto_mutex is held, or it can be called from ->run(), in which
6088 * case ofproto_mutex is not held. But only in the latter case can we
6089 * arrive here, so we can safely take ofproto_mutex now. */
6090 ovs_mutex_lock(&ofproto_mutex);
6091 ovs_assert(op->rule->pending == op);
6092 ofopgroup_complete(group);
6093 ovs_mutex_unlock(&ofproto_mutex);
6098 pick_datapath_id(const struct ofproto *ofproto)
6100 const struct ofport *port;
6102 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6104 uint8_t ea[ETH_ADDR_LEN];
6107 error = netdev_get_etheraddr(port->netdev, ea);
6109 return eth_addr_to_uint64(ea);
6111 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6112 ofproto->name, netdev_get_name(port->netdev),
6113 ovs_strerror(error));
6115 return ofproto->fallback_dpid;
6119 pick_fallback_dpid(void)
6121 uint8_t ea[ETH_ADDR_LEN];
6122 eth_addr_nicira_random(ea);
6123 return eth_addr_to_uint64(ea);
6126 /* Table overflow policy. */
6128 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6129 * to NULL if the table is not configured to evict rules or if the table
6130 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6131 * or with no timeouts are not evictable.) */
6133 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6134 OVS_REQUIRES(ofproto_mutex)
6136 struct eviction_group *evg;
6139 if (!table->eviction_fields) {
6143 /* In the common case, the outer and inner loops here will each be entered
6146 * - The inner loop normally "return"s in its first iteration. If the
6147 * eviction group has any evictable rules, then it always returns in
6150 * - The outer loop only iterates more than once if the largest eviction
6151 * group has no evictable rules.
6153 * - The outer loop can exit only if table's 'max_flows' is all filled up
6154 * by unevictable rules. */
6155 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6158 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6167 /* Searches 'ofproto' for tables that have more flows than their configured
6168 * maximum and that have flow eviction enabled, and evicts as many flows as
6169 * necessary and currently feasible from them.
6171 * This triggers only when an OpenFlow table has N flows in it and then the
6172 * client configures a maximum number of flows less than N. */
6174 ofproto_evict(struct ofproto *ofproto)
6176 struct oftable *table;
6178 ovs_mutex_lock(&ofproto_mutex);
6179 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6180 evict_rules_from_table(ofproto, table, 0);
6182 ovs_mutex_unlock(&ofproto_mutex);
6185 /* Eviction groups. */
6187 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6188 * rules. The priority contains low-order random bits to ensure that eviction
6189 * groups with the same number of rules are prioritized randomly. */
6191 eviction_group_priority(size_t n_rules)
6193 uint16_t size = MIN(UINT16_MAX, n_rules);
6194 return (size << 16) | random_uint16();
6197 /* Updates 'evg', an eviction_group within 'table', following a change that
6198 * adds or removes rules in 'evg'. */
6200 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6201 OVS_REQUIRES(ofproto_mutex)
6203 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6204 eviction_group_priority(heap_count(&evg->rules)));
6207 /* Destroys 'evg', an eviction_group within 'table':
6209 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6210 * rules themselves, just removes them from the eviction group.)
6212 * - Removes 'evg' from 'table'.
6216 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6217 OVS_REQUIRES(ofproto_mutex)
6219 while (!heap_is_empty(&evg->rules)) {
6222 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6223 rule->eviction_group = NULL;
6225 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6226 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6227 heap_destroy(&evg->rules);
6231 /* Removes 'rule' from its eviction group, if any. */
6233 eviction_group_remove_rule(struct rule *rule)
6234 OVS_REQUIRES(ofproto_mutex)
6236 if (rule->eviction_group) {
6237 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6238 struct eviction_group *evg = rule->eviction_group;
6240 rule->eviction_group = NULL;
6241 heap_remove(&evg->rules, &rule->evg_node);
6242 if (heap_is_empty(&evg->rules)) {
6243 eviction_group_destroy(table, evg);
6245 eviction_group_resized(table, evg);
6250 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6251 * returns the hash value. */
6253 eviction_group_hash_rule(struct rule *rule)
6254 OVS_REQUIRES(ofproto_mutex)
6256 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6257 const struct mf_subfield *sf;
6261 hash = table->eviction_group_id_basis;
6262 miniflow_expand(&rule->cr.match.flow, &flow);
6263 for (sf = table->eviction_fields;
6264 sf < &table->eviction_fields[table->n_eviction_fields];
6267 if (mf_are_prereqs_ok(sf->field, &flow)) {
6268 union mf_value value;
6270 mf_get_value(sf->field, &flow, &value);
6272 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6274 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6275 unsigned int start = sf->ofs + sf->n_bits;
6276 bitwise_zero(&value, sf->field->n_bytes, start,
6277 sf->field->n_bytes * 8 - start);
6279 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6281 hash = hash_int(hash, 0);
6288 /* Returns an eviction group within 'table' with the given 'id', creating one
6290 static struct eviction_group *
6291 eviction_group_find(struct oftable *table, uint32_t id)
6292 OVS_REQUIRES(ofproto_mutex)
6294 struct eviction_group *evg;
6296 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6300 evg = xmalloc(sizeof *evg);
6301 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6302 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6303 eviction_group_priority(0));
6304 heap_init(&evg->rules);
6309 /* Returns an eviction priority for 'rule'. The return value should be
6310 * interpreted so that higher priorities make a rule more attractive candidates
6313 rule_eviction_priority(struct rule *rule)
6314 OVS_REQUIRES(ofproto_mutex)
6316 long long int hard_expiration;
6317 long long int idle_expiration;
6318 long long int expiration;
6319 uint32_t expiration_offset;
6321 /* Calculate time of expiration. */
6322 ovs_mutex_lock(&rule->mutex);
6323 hard_expiration = (rule->hard_timeout
6324 ? rule->modified + rule->hard_timeout * 1000
6326 idle_expiration = (rule->idle_timeout
6327 ? rule->used + rule->idle_timeout * 1000
6329 expiration = MIN(hard_expiration, idle_expiration);
6330 ovs_mutex_unlock(&rule->mutex);
6331 if (expiration == LLONG_MAX) {
6335 /* Calculate the time of expiration as a number of (approximate) seconds
6336 * after program startup.
6338 * This should work OK for program runs that last UINT32_MAX seconds or
6339 * less. Therefore, please restart OVS at least once every 136 years. */
6340 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6342 /* Invert the expiration offset because we're using a max-heap. */
6343 return UINT32_MAX - expiration_offset;
6346 /* Adds 'rule' to an appropriate eviction group for its oftable's
6347 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6348 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6351 * The caller must ensure that 'rule' is not already in an eviction group. */
6353 eviction_group_add_rule(struct rule *rule)
6354 OVS_REQUIRES(ofproto_mutex)
6356 struct ofproto *ofproto = rule->ofproto;
6357 struct oftable *table = &ofproto->tables[rule->table_id];
6360 ovs_mutex_lock(&rule->mutex);
6361 has_timeout = rule->hard_timeout || rule->idle_timeout;
6362 ovs_mutex_unlock(&rule->mutex);
6364 if (table->eviction_fields && has_timeout) {
6365 struct eviction_group *evg;
6367 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6369 rule->eviction_group = evg;
6370 heap_insert(&evg->rules, &rule->evg_node,
6371 rule_eviction_priority(rule));
6372 eviction_group_resized(table, evg);
6378 /* Initializes 'table'. */
6380 oftable_init(struct oftable *table)
6382 memset(table, 0, sizeof *table);
6383 classifier_init(&table->cls);
6384 table->max_flows = UINT_MAX;
6387 /* Destroys 'table', including its classifier and eviction groups.
6389 * The caller is responsible for freeing 'table' itself. */
6391 oftable_destroy(struct oftable *table)
6393 ovs_rwlock_rdlock(&table->cls.rwlock);
6394 ovs_assert(classifier_is_empty(&table->cls));
6395 ovs_rwlock_unlock(&table->cls.rwlock);
6396 oftable_disable_eviction(table);
6397 classifier_destroy(&table->cls);
6401 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6402 * string, then 'table' will use its default name.
6404 * This only affects the name exposed for a table exposed through the OpenFlow
6405 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6407 oftable_set_name(struct oftable *table, const char *name)
6409 if (name && name[0]) {
6410 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6411 if (!table->name || strncmp(name, table->name, len)) {
6413 table->name = xmemdup0(name, len);
6421 /* oftables support a choice of two policies when adding a rule would cause the
6422 * number of flows in the table to exceed the configured maximum number: either
6423 * they can refuse to add the new flow or they can evict some existing flow.
6424 * This function configures the former policy on 'table'. */
6426 oftable_disable_eviction(struct oftable *table)
6427 OVS_REQUIRES(ofproto_mutex)
6429 if (table->eviction_fields) {
6430 struct eviction_group *evg, *next;
6432 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6433 &table->eviction_groups_by_id) {
6434 eviction_group_destroy(table, evg);
6436 hmap_destroy(&table->eviction_groups_by_id);
6437 heap_destroy(&table->eviction_groups_by_size);
6439 free(table->eviction_fields);
6440 table->eviction_fields = NULL;
6441 table->n_eviction_fields = 0;
6445 /* oftables support a choice of two policies when adding a rule would cause the
6446 * number of flows in the table to exceed the configured maximum number: either
6447 * they can refuse to add the new flow or they can evict some existing flow.
6448 * This function configures the latter policy on 'table', with fairness based
6449 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6450 * 'n_fields' as 0 disables fairness.) */
6452 oftable_enable_eviction(struct oftable *table,
6453 const struct mf_subfield *fields, size_t n_fields)
6454 OVS_REQUIRES(ofproto_mutex)
6456 struct cls_cursor cursor;
6459 if (table->eviction_fields
6460 && n_fields == table->n_eviction_fields
6462 || !memcmp(fields, table->eviction_fields,
6463 n_fields * sizeof *fields))) {
6468 oftable_disable_eviction(table);
6470 table->n_eviction_fields = n_fields;
6471 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6473 table->eviction_group_id_basis = random_uint32();
6474 hmap_init(&table->eviction_groups_by_id);
6475 heap_init(&table->eviction_groups_by_size);
6477 ovs_rwlock_rdlock(&table->cls.rwlock);
6478 cls_cursor_init(&cursor, &table->cls, NULL);
6479 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6480 eviction_group_add_rule(rule);
6482 ovs_rwlock_unlock(&table->cls.rwlock);
6485 /* Removes 'rule' from the oftable that contains it. */
6487 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6488 OVS_REQUIRES(ofproto_mutex)
6490 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6492 ovs_rwlock_wrlock(&cls->rwlock);
6493 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6494 ovs_rwlock_unlock(&cls->rwlock);
6496 cookies_remove(ofproto, rule);
6498 eviction_group_remove_rule(rule);
6499 if (!list_is_empty(&rule->expirable)) {
6500 list_remove(&rule->expirable);
6502 if (!list_is_empty(&rule->meter_list_node)) {
6503 list_remove(&rule->meter_list_node);
6504 list_init(&rule->meter_list_node);
6509 oftable_remove_rule(struct rule *rule)
6510 OVS_REQUIRES(ofproto_mutex)
6512 oftable_remove_rule__(rule->ofproto, rule);
6515 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6516 * the same cls_rule. */
6518 oftable_insert_rule(struct rule *rule)
6519 OVS_REQUIRES(ofproto_mutex)
6521 struct ofproto *ofproto = rule->ofproto;
6522 struct oftable *table = &ofproto->tables[rule->table_id];
6525 ovs_mutex_lock(&rule->mutex);
6526 may_expire = rule->hard_timeout || rule->idle_timeout;
6527 ovs_mutex_unlock(&rule->mutex);
6530 list_insert(&ofproto->expirable, &rule->expirable);
6533 cookies_insert(ofproto, rule);
6535 if (rule->actions->meter_id) {
6536 struct meter *meter = ofproto->meters[rule->actions->meter_id];
6537 list_insert(&meter->rules, &rule->meter_list_node);
6539 ovs_rwlock_wrlock(&table->cls.rwlock);
6540 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6541 ovs_rwlock_unlock(&table->cls.rwlock);
6542 eviction_group_add_rule(rule);
6545 /* unixctl commands. */
6548 ofproto_lookup(const char *name)
6550 struct ofproto *ofproto;
6552 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6554 if (!strcmp(ofproto->name, name)) {
6562 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6563 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6565 struct ofproto *ofproto;
6569 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6570 ds_put_format(&results, "%s\n", ofproto->name);
6572 unixctl_command_reply(conn, ds_cstr(&results));
6573 ds_destroy(&results);
6577 ofproto_unixctl_init(void)
6579 static bool registered;
6585 unixctl_command_register("ofproto/list", "", 0, 0,
6586 ofproto_unixctl_list, NULL);
6589 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6591 * This is deprecated. It is only for compatibility with broken device drivers
6592 * in old versions of Linux that do not properly support VLANs when VLAN
6593 * devices are not used. When broken device drivers are no longer in
6594 * widespread use, we will delete these interfaces. */
6596 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6597 * (exactly) by an OpenFlow rule in 'ofproto'. */
6599 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6601 const struct oftable *oftable;
6603 free(ofproto->vlan_bitmap);
6604 ofproto->vlan_bitmap = bitmap_allocate(4096);
6605 ofproto->vlans_changed = false;
6607 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6608 const struct cls_table *table;
6610 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6611 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
6612 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6613 const struct cls_rule *rule;
6615 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6616 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6617 bitmap_set1(vlan_bitmap, vid);
6618 bitmap_set1(ofproto->vlan_bitmap, vid);
6622 ovs_rwlock_unlock(&oftable->cls.rwlock);
6626 /* Returns true if new VLANs have come into use by the flow table since the
6627 * last call to ofproto_get_vlan_usage().
6629 * We don't track when old VLANs stop being used. */
6631 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6633 return ofproto->vlans_changed;
6636 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6637 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6638 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6639 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6640 * then the VLAN device is un-enslaved. */
6642 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6643 ofp_port_t realdev_ofp_port, int vid)
6645 struct ofport *ofport;
6648 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6650 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6652 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6653 ofproto->name, vlandev_ofp_port);
6657 if (!ofproto->ofproto_class->set_realdev) {
6658 if (!vlandev_ofp_port) {
6661 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6665 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6667 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6668 ofproto->name, vlandev_ofp_port,
6669 netdev_get_name(ofport->netdev), ovs_strerror(error));