2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
37 #include "ofp-print.h"
40 #include "ofproto-provider.h"
41 #include "openflow/nicira-ext.h"
42 #include "openflow/openflow.h"
46 #include "poll-loop.h"
52 #include "unaligned.h"
56 VLOG_DEFINE_THIS_MODULE(ofproto);
58 COVERAGE_DEFINE(ofproto_error);
59 COVERAGE_DEFINE(ofproto_flush);
60 COVERAGE_DEFINE(ofproto_no_packet_in);
61 COVERAGE_DEFINE(ofproto_packet_out);
62 COVERAGE_DEFINE(ofproto_queue_req);
63 COVERAGE_DEFINE(ofproto_recv_openflow);
64 COVERAGE_DEFINE(ofproto_reinit_ports);
65 COVERAGE_DEFINE(ofproto_uninstallable);
66 COVERAGE_DEFINE(ofproto_update_port);
69 S_OPENFLOW, /* Processing OpenFlow commands. */
70 S_EVICT, /* Evicting flows from over-limit tables. */
71 S_FLUSH, /* Deleting all flow table rules. */
74 enum ofoperation_type {
80 /* A single OpenFlow request can execute any number of operations. The
81 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
82 * ofconn to which an error reply should be sent if necessary.
84 * ofproto initiates some operations internally. These operations are still
85 * assigned to groups but will not have an associated ofconn. */
87 struct ofproto *ofproto; /* Owning ofproto. */
88 struct list ofproto_node; /* In ofproto's "pending" list. */
89 struct list ops; /* List of "struct ofoperation"s. */
90 int n_running; /* Number of ops still pending. */
92 /* Data needed to send OpenFlow reply on failure or to send a buffered
95 * If list_is_empty(ofconn_node) then this ofopgroup never had an
96 * associated ofconn or its ofconn's connection dropped after it initiated
97 * the operation. In the latter case 'ofconn' is a wild pointer that
98 * refers to freed memory, so the 'ofconn' member must be used only if
99 * !list_is_empty(ofconn_node).
101 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
102 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
103 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
104 uint32_t buffer_id; /* Buffer id from original request. */
107 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
108 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
109 const struct ofp_header *,
111 static void ofopgroup_submit(struct ofopgroup *);
112 static void ofopgroup_complete(struct ofopgroup *);
114 /* A single flow table operation. */
116 struct ofopgroup *group; /* Owning group. */
117 struct list group_node; /* In ofopgroup's "ops" list. */
118 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
119 struct rule *rule; /* Rule being operated upon. */
120 enum ofoperation_type type; /* Type of operation. */
122 /* OFOPERATION_ADD. */
123 struct rule *victim; /* Rule being replaced, if any.. */
125 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
126 struct ofpact *ofpacts;
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 enum ofperr error; /* 0 if no error. */
136 static struct ofoperation *ofoperation_create(struct ofopgroup *,
138 enum ofoperation_type,
139 enum ofp_flow_removed_reason);
140 static void ofoperation_destroy(struct ofoperation *);
143 static void oftable_init(struct oftable *);
144 static void oftable_destroy(struct oftable *);
146 static void oftable_set_name(struct oftable *, const char *name);
148 static void oftable_disable_eviction(struct oftable *);
149 static void oftable_enable_eviction(struct oftable *,
150 const struct mf_subfield *fields,
153 static void oftable_remove_rule(struct rule *);
154 static struct rule *oftable_replace_rule(struct rule *);
155 static void oftable_substitute_rule(struct rule *old, struct rule *new);
157 /* A set of rules within a single OpenFlow table (oftable) that have the same
158 * values for the oftable's eviction_fields. A rule to be evicted, when one is
159 * needed, is taken from the eviction group that contains the greatest number
162 * An oftable owns any number of eviction groups, each of which contains any
165 * Membership in an eviction group is imprecise, based on the hash of the
166 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
167 * That is, if two rules have different eviction_fields, but those
168 * eviction_fields hash to the same value, then they will belong to the same
169 * eviction_group anyway.
171 * (When eviction is not enabled on an oftable, we don't track any eviction
172 * groups, to save time and space.) */
173 struct eviction_group {
174 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
175 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
176 struct heap rules; /* Contains "struct rule"s. */
179 static struct rule *choose_rule_to_evict(struct oftable *);
180 static void ofproto_evict(struct ofproto *);
181 static uint32_t rule_eviction_priority(struct rule *);
184 static void ofport_destroy__(struct ofport *);
185 static void ofport_destroy(struct ofport *);
187 static void update_port(struct ofproto *, const char *devname);
188 static int init_ports(struct ofproto *);
189 static void reinit_ports(struct ofproto *);
192 static void ofproto_rule_destroy__(struct rule *);
193 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
194 static bool rule_is_modifiable(const struct rule *);
197 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
198 const struct ofputil_flow_mod *,
199 const struct ofp_header *);
200 static void delete_flow__(struct rule *, struct ofopgroup *);
201 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
202 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
203 const struct ofputil_flow_mod *,
204 const struct ofp_header *);
207 static uint64_t pick_datapath_id(const struct ofproto *);
208 static uint64_t pick_fallback_dpid(void);
209 static void ofproto_destroy__(struct ofproto *);
210 static void update_mtu(struct ofproto *, struct ofport *);
213 static void ofproto_unixctl_init(void);
215 /* All registered ofproto classes, in probe order. */
216 static const struct ofproto_class **ofproto_classes;
217 static size_t n_ofproto_classes;
218 static size_t allocated_ofproto_classes;
220 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
221 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
223 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
226 ofproto_initialize(void)
232 ofproto_class_register(&ofproto_dpif_class);
236 /* 'type' should be a normalized datapath type, as returned by
237 * ofproto_normalize_type(). Returns the corresponding ofproto_class
238 * structure, or a null pointer if there is none registered for 'type'. */
239 static const struct ofproto_class *
240 ofproto_class_find__(const char *type)
244 ofproto_initialize();
245 for (i = 0; i < n_ofproto_classes; i++) {
246 const struct ofproto_class *class = ofproto_classes[i];
251 class->enumerate_types(&types);
252 found = sset_contains(&types, type);
253 sset_destroy(&types);
259 VLOG_WARN("unknown datapath type %s", type);
263 /* Registers a new ofproto class. After successful registration, new ofprotos
264 * of that type can be created using ofproto_create(). */
266 ofproto_class_register(const struct ofproto_class *new_class)
270 for (i = 0; i < n_ofproto_classes; i++) {
271 if (ofproto_classes[i] == new_class) {
276 if (n_ofproto_classes >= allocated_ofproto_classes) {
277 ofproto_classes = x2nrealloc(ofproto_classes,
278 &allocated_ofproto_classes,
279 sizeof *ofproto_classes);
281 ofproto_classes[n_ofproto_classes++] = new_class;
285 /* Unregisters a datapath provider. 'type' must have been previously
286 * registered and not currently be in use by any ofprotos. After
287 * unregistration new datapaths of that type cannot be opened using
288 * ofproto_create(). */
290 ofproto_class_unregister(const struct ofproto_class *class)
294 for (i = 0; i < n_ofproto_classes; i++) {
295 if (ofproto_classes[i] == class) {
296 for (i++; i < n_ofproto_classes; i++) {
297 ofproto_classes[i - 1] = ofproto_classes[i];
303 VLOG_WARN("attempted to unregister an ofproto class that is not "
308 /* Clears 'types' and enumerates all registered ofproto types into it. The
309 * caller must first initialize the sset. */
311 ofproto_enumerate_types(struct sset *types)
315 ofproto_initialize();
316 for (i = 0; i < n_ofproto_classes; i++) {
317 ofproto_classes[i]->enumerate_types(types);
321 /* Returns the fully spelled out name for the given ofproto 'type'.
323 * Normalized type string can be compared with strcmp(). Unnormalized type
324 * string might be the same even if they have different spellings. */
326 ofproto_normalize_type(const char *type)
328 return type && type[0] ? type : "system";
331 /* Clears 'names' and enumerates the names of all known created ofprotos with
332 * the given 'type'. The caller must first initialize the sset. Returns 0 if
333 * successful, otherwise a positive errno value.
335 * Some kinds of datapaths might not be practically enumerable. This is not
336 * considered an error. */
338 ofproto_enumerate_names(const char *type, struct sset *names)
340 const struct ofproto_class *class = ofproto_class_find__(type);
341 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
345 ofproto_create(const char *datapath_name, const char *datapath_type,
346 struct ofproto **ofprotop)
348 const struct ofproto_class *class;
349 struct ofproto *ofproto;
354 ofproto_initialize();
355 ofproto_unixctl_init();
357 datapath_type = ofproto_normalize_type(datapath_type);
358 class = ofproto_class_find__(datapath_type);
360 VLOG_WARN("could not create datapath %s of unknown type %s",
361 datapath_name, datapath_type);
365 ofproto = class->alloc();
367 VLOG_ERR("failed to allocate datapath %s of type %s",
368 datapath_name, datapath_type);
373 memset(ofproto, 0, sizeof *ofproto);
374 ofproto->ofproto_class = class;
375 ofproto->name = xstrdup(datapath_name);
376 ofproto->type = xstrdup(datapath_type);
377 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
378 hash_string(ofproto->name, 0));
379 ofproto->datapath_id = 0;
380 ofproto_set_flow_eviction_threshold(ofproto,
381 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
382 ofproto->forward_bpdu = false;
383 ofproto->fallback_dpid = pick_fallback_dpid();
384 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
385 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
386 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
387 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
388 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
389 ofproto->frag_handling = OFPC_FRAG_NORMAL;
390 hmap_init(&ofproto->ports);
391 shash_init(&ofproto->port_by_name);
392 ofproto->tables = NULL;
393 ofproto->n_tables = 0;
394 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
395 ofproto->state = S_OPENFLOW;
396 list_init(&ofproto->pending);
397 ofproto->n_pending = 0;
398 hmap_init(&ofproto->deletions);
399 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
400 ofproto->first_op = ofproto->last_op = LLONG_MIN;
401 ofproto->next_op_report = LLONG_MAX;
402 ofproto->op_backoff = LLONG_MIN;
403 ofproto->vlan_bitmap = NULL;
404 ofproto->vlans_changed = false;
405 ofproto->min_mtu = INT_MAX;
407 error = ofproto->ofproto_class->construct(ofproto);
409 VLOG_ERR("failed to open datapath %s: %s",
410 datapath_name, strerror(error));
411 ofproto_destroy__(ofproto);
415 assert(ofproto->n_tables);
417 ofproto->datapath_id = pick_datapath_id(ofproto);
425 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
427 struct oftable *table;
429 assert(!ofproto->n_tables);
430 assert(n_tables >= 1 && n_tables <= 255);
432 ofproto->n_tables = n_tables;
433 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
434 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
440 ofproto_get_datapath_id(const struct ofproto *ofproto)
442 return ofproto->datapath_id;
446 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
448 uint64_t old_dpid = p->datapath_id;
449 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
450 if (p->datapath_id != old_dpid) {
451 /* Force all active connections to reconnect, since there is no way to
452 * notify a controller that the datapath ID has changed. */
453 ofproto_reconnect_controllers(p);
458 ofproto_set_controllers(struct ofproto *p,
459 const struct ofproto_controller *controllers,
460 size_t n_controllers)
462 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
466 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
468 connmgr_set_fail_mode(p->connmgr, fail_mode);
471 /* Drops the connections between 'ofproto' and all of its controllers, forcing
472 * them to reconnect. */
474 ofproto_reconnect_controllers(struct ofproto *ofproto)
476 connmgr_reconnect(ofproto->connmgr);
479 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
480 * in-band control should guarantee access, in the same way that in-band
481 * control guarantees access to OpenFlow controllers. */
483 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
484 const struct sockaddr_in *extras, size_t n)
486 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
489 /* Sets the OpenFlow queue used by flows set up by in-band control on
490 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
491 * flows will use the default queue. */
493 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
495 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
498 /* Sets the number of flows at which eviction from the kernel flow table
501 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
503 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
504 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
506 ofproto->flow_eviction_threshold = threshold;
510 /* If forward_bpdu is true, the NORMAL action will forward frames with
511 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
512 * the NORMAL action will drop these frames. */
514 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
516 bool old_val = ofproto->forward_bpdu;
517 ofproto->forward_bpdu = forward_bpdu;
518 if (old_val != ofproto->forward_bpdu) {
519 if (ofproto->ofproto_class->forward_bpdu_changed) {
520 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
525 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
526 * 'idle_time', in seconds. */
528 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
530 if (ofproto->ofproto_class->set_mac_idle_time) {
531 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
536 ofproto_set_desc(struct ofproto *p,
537 const char *mfr_desc, const char *hw_desc,
538 const char *sw_desc, const char *serial_desc,
541 struct ofp_desc_stats *ods;
544 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
545 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
546 p->name, sizeof ods->mfr_desc);
549 p->mfr_desc = xstrdup(mfr_desc);
552 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
553 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
554 p->name, sizeof ods->hw_desc);
557 p->hw_desc = xstrdup(hw_desc);
560 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
561 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
562 p->name, sizeof ods->sw_desc);
565 p->sw_desc = xstrdup(sw_desc);
568 if (strlen(serial_desc) >= sizeof ods->serial_num) {
569 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
570 "bytes", p->name, sizeof ods->serial_num);
572 free(p->serial_desc);
573 p->serial_desc = xstrdup(serial_desc);
576 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
577 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
578 p->name, sizeof ods->dp_desc);
581 p->dp_desc = xstrdup(dp_desc);
586 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
588 return connmgr_set_snoops(ofproto->connmgr, snoops);
592 ofproto_set_netflow(struct ofproto *ofproto,
593 const struct netflow_options *nf_options)
595 if (nf_options && sset_is_empty(&nf_options->collectors)) {
599 if (ofproto->ofproto_class->set_netflow) {
600 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
602 return nf_options ? EOPNOTSUPP : 0;
607 ofproto_set_sflow(struct ofproto *ofproto,
608 const struct ofproto_sflow_options *oso)
610 if (oso && sset_is_empty(&oso->targets)) {
614 if (ofproto->ofproto_class->set_sflow) {
615 return ofproto->ofproto_class->set_sflow(ofproto, oso);
617 return oso ? EOPNOTSUPP : 0;
621 /* Spanning Tree Protocol (STP) configuration. */
623 /* Configures STP on 'ofproto' using the settings defined in 's'. If
624 * 's' is NULL, disables STP.
626 * Returns 0 if successful, otherwise a positive errno value. */
628 ofproto_set_stp(struct ofproto *ofproto,
629 const struct ofproto_stp_settings *s)
631 return (ofproto->ofproto_class->set_stp
632 ? ofproto->ofproto_class->set_stp(ofproto, s)
636 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
637 * 'enabled' member of 's' is false, then the other members are not
640 * Returns 0 if successful, otherwise a positive errno value. */
642 ofproto_get_stp_status(struct ofproto *ofproto,
643 struct ofproto_stp_status *s)
645 return (ofproto->ofproto_class->get_stp_status
646 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
650 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
651 * in 's'. The caller is responsible for assigning STP port numbers
652 * (using the 'port_num' member in the range of 1 through 255, inclusive)
653 * and ensuring there are no duplicates. If the 's' is NULL, then STP
654 * is disabled on the port.
656 * Returns 0 if successful, otherwise a positive errno value.*/
658 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
659 const struct ofproto_port_stp_settings *s)
661 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
663 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
664 ofproto->name, ofp_port);
668 return (ofproto->ofproto_class->set_stp_port
669 ? ofproto->ofproto_class->set_stp_port(ofport, s)
673 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
674 * 's'. If the 'enabled' member in 's' is false, then the other members
675 * are not meaningful.
677 * Returns 0 if successful, otherwise a positive errno value.*/
679 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
680 struct ofproto_port_stp_status *s)
682 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
684 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
685 "port %"PRIu16, ofproto->name, ofp_port);
689 return (ofproto->ofproto_class->get_stp_port_status
690 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
694 /* Queue DSCP configuration. */
696 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
697 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
698 * to implement QoS. Instead, it is used to implement features which require
699 * knowledge of what queues exist on a port, and some basic information about
702 * Returns 0 if successful, otherwise a positive errno value. */
704 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
705 const struct ofproto_port_queue *queues,
708 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
711 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
712 ofproto->name, ofp_port);
716 return (ofproto->ofproto_class->set_queues
717 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
721 /* Connectivity Fault Management configuration. */
723 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
725 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
727 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
728 if (ofport && ofproto->ofproto_class->set_cfm) {
729 ofproto->ofproto_class->set_cfm(ofport, NULL);
733 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
734 * basic configuration from the configuration members in 'cfm', and the remote
735 * maintenance point ID from remote_mpid. Ignores the statistics members of
738 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
740 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
741 const struct cfm_settings *s)
743 struct ofport *ofport;
746 ofport = ofproto_get_port(ofproto, ofp_port);
748 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
749 ofproto->name, ofp_port);
753 /* XXX: For configuration simplicity, we only support one remote_mpid
754 * outside of the CFM module. It's not clear if this is the correct long
755 * term solution or not. */
756 error = (ofproto->ofproto_class->set_cfm
757 ? ofproto->ofproto_class->set_cfm(ofport, s)
760 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
761 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
766 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
767 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
768 * 0 if LACP partner information is not current (generally indicating a
769 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
771 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
773 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
774 return (ofport && ofproto->ofproto_class->port_is_lacp_current
775 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
781 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
782 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
783 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
784 * configuration plus, if there is more than one slave, a bonding
787 * If 'aux' is already registered then this function updates its configuration
788 * to 's'. Otherwise, this function registers a new bundle.
790 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
793 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
794 const struct ofproto_bundle_settings *s)
796 return (ofproto->ofproto_class->bundle_set
797 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
801 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
802 * If no such bundle has been registered, this has no effect. */
804 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
806 return ofproto_bundle_register(ofproto, aux, NULL);
810 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
811 * If 'aux' is already registered then this function updates its configuration
812 * to 's'. Otherwise, this function registers a new mirror. */
814 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
815 const struct ofproto_mirror_settings *s)
817 return (ofproto->ofproto_class->mirror_set
818 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
822 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
823 * If no mirror has been registered, this has no effect. */
825 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
827 return ofproto_mirror_register(ofproto, aux, NULL);
830 /* Retrieves statistics from mirror associated with client data pointer
831 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
832 * 'bytes', respectively. If a particular counters is not supported,
833 * the appropriate argument is set to UINT64_MAX. */
835 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
836 uint64_t *packets, uint64_t *bytes)
838 if (!ofproto->ofproto_class->mirror_get_stats) {
839 *packets = *bytes = UINT64_MAX;
843 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
847 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
848 * which all packets are flooded, instead of using MAC learning. If
849 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
851 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
854 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
856 return (ofproto->ofproto_class->set_flood_vlans
857 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
861 /* Returns true if 'aux' is a registered bundle that is currently in use as the
862 * output for a mirror. */
864 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
866 return (ofproto->ofproto_class->is_mirror_output_bundle
867 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
871 /* Configuration of OpenFlow tables. */
873 /* Returns the number of OpenFlow tables in 'ofproto'. */
875 ofproto_get_n_tables(const struct ofproto *ofproto)
877 return ofproto->n_tables;
880 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
881 * settings from 's'. 'table_id' must be in the range 0 through the number of
882 * OpenFlow tables in 'ofproto' minus 1, inclusive.
884 * For read-only tables, only the name may be configured. */
886 ofproto_configure_table(struct ofproto *ofproto, int table_id,
887 const struct ofproto_table_settings *s)
889 struct oftable *table;
891 assert(table_id >= 0 && table_id < ofproto->n_tables);
892 table = &ofproto->tables[table_id];
894 oftable_set_name(table, s->name);
896 if (table->flags & OFTABLE_READONLY) {
901 oftable_enable_eviction(table, s->groups, s->n_groups);
903 oftable_disable_eviction(table);
906 table->max_flows = s->max_flows;
907 if (classifier_count(&table->cls) > table->max_flows
908 && table->eviction_fields) {
909 /* 'table' contains more flows than allowed. We might not be able to
910 * evict them right away because of the asynchronous nature of flow
911 * table changes. Schedule eviction for later. */
912 switch (ofproto->state) {
914 ofproto->state = S_EVICT;
918 /* We're already deleting flows, nothing more to do. */
925 ofproto_has_snoops(const struct ofproto *ofproto)
927 return connmgr_has_snoops(ofproto->connmgr);
931 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
933 connmgr_get_snoops(ofproto->connmgr, snoops);
937 ofproto_flush__(struct ofproto *ofproto)
939 struct ofopgroup *group;
940 struct oftable *table;
942 if (ofproto->ofproto_class->flush) {
943 ofproto->ofproto_class->flush(ofproto);
946 group = ofopgroup_create_unattached(ofproto);
947 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
948 struct rule *rule, *next_rule;
949 struct cls_cursor cursor;
951 if (table->flags & OFTABLE_HIDDEN) {
955 cls_cursor_init(&cursor, &table->cls, NULL);
956 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
957 if (!rule->pending) {
958 ofoperation_create(group, rule, OFOPERATION_DELETE,
960 oftable_remove_rule(rule);
961 ofproto->ofproto_class->rule_destruct(rule);
965 ofopgroup_submit(group);
969 ofproto_destroy__(struct ofproto *ofproto)
971 struct oftable *table;
973 assert(list_is_empty(&ofproto->pending));
974 assert(!ofproto->n_pending);
976 connmgr_destroy(ofproto->connmgr);
978 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
981 free(ofproto->mfr_desc);
982 free(ofproto->hw_desc);
983 free(ofproto->sw_desc);
984 free(ofproto->serial_desc);
985 free(ofproto->dp_desc);
986 hmap_destroy(&ofproto->ports);
987 shash_destroy(&ofproto->port_by_name);
989 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
990 oftable_destroy(table);
992 free(ofproto->tables);
994 hmap_destroy(&ofproto->deletions);
996 free(ofproto->vlan_bitmap);
998 ofproto->ofproto_class->dealloc(ofproto);
1002 ofproto_destroy(struct ofproto *p)
1004 struct ofport *ofport, *next_ofport;
1011 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1012 ofport_destroy(ofport);
1015 p->ofproto_class->destruct(p);
1016 ofproto_destroy__(p);
1019 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1020 * kernel datapath, for example, this destroys the datapath in the kernel, and
1021 * with the netdev-based datapath, it tears down the data structures that
1022 * represent the datapath.
1024 * The datapath should not be currently open as an ofproto. */
1026 ofproto_delete(const char *name, const char *type)
1028 const struct ofproto_class *class = ofproto_class_find__(type);
1029 return (!class ? EAFNOSUPPORT
1030 : !class->del ? EACCES
1031 : class->del(type, name));
1035 process_port_change(struct ofproto *ofproto, int error, char *devname)
1037 if (error == ENOBUFS) {
1038 reinit_ports(ofproto);
1039 } else if (!error) {
1040 update_port(ofproto, devname);
1046 ofproto_run(struct ofproto *p)
1048 struct sset changed_netdevs;
1049 const char *changed_netdev;
1050 struct ofport *ofport;
1053 error = p->ofproto_class->run(p);
1054 if (error && error != EAGAIN) {
1055 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1058 if (p->ofproto_class->port_poll) {
1061 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1062 process_port_change(p, error, devname);
1066 /* Update OpenFlow port status for any port whose netdev has changed.
1068 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1069 * destroyed, so it's not safe to update ports directly from the
1070 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1071 * need this two-phase approach. */
1072 sset_init(&changed_netdevs);
1073 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1074 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1075 if (ofport->change_seq != change_seq) {
1076 ofport->change_seq = change_seq;
1077 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1080 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1081 update_port(p, changed_netdev);
1083 sset_destroy(&changed_netdevs);
1087 connmgr_run(p->connmgr, handle_openflow);
1091 connmgr_run(p->connmgr, NULL);
1093 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1094 p->state = S_OPENFLOW;
1099 connmgr_run(p->connmgr, NULL);
1101 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1102 connmgr_flushed(p->connmgr);
1103 p->state = S_OPENFLOW;
1111 if (time_msec() >= p->next_op_report) {
1112 long long int ago = (time_msec() - p->first_op) / 1000;
1113 long long int interval = (p->last_op - p->first_op) / 1000;
1117 ds_put_format(&s, "%d flow_mods ",
1118 p->n_add + p->n_delete + p->n_modify);
1119 if (interval == ago) {
1120 ds_put_format(&s, "in the last %lld s", ago);
1121 } else if (interval) {
1122 ds_put_format(&s, "in the %lld s starting %lld s ago",
1125 ds_put_format(&s, "%lld s ago", ago);
1128 ds_put_cstr(&s, " (");
1130 ds_put_format(&s, "%d adds, ", p->n_add);
1133 ds_put_format(&s, "%d deletes, ", p->n_delete);
1136 ds_put_format(&s, "%d modifications, ", p->n_modify);
1139 ds_put_char(&s, ')');
1141 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1144 p->n_add = p->n_delete = p->n_modify = 0;
1145 p->next_op_report = LLONG_MAX;
1151 /* Performs periodic activity required by 'ofproto' that needs to be done
1152 * with the least possible latency.
1154 * It makes sense to call this function a couple of times per poll loop, to
1155 * provide a significant performance boost on some benchmarks with the
1156 * ofproto-dpif implementation. */
1158 ofproto_run_fast(struct ofproto *p)
1162 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1163 if (error && error != EAGAIN) {
1164 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1165 p->name, strerror(error));
1171 ofproto_wait(struct ofproto *p)
1173 struct ofport *ofport;
1175 p->ofproto_class->wait(p);
1176 if (p->ofproto_class->port_poll_wait) {
1177 p->ofproto_class->port_poll_wait(p);
1180 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1181 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1182 poll_immediate_wake();
1188 connmgr_wait(p->connmgr, true);
1193 connmgr_wait(p->connmgr, false);
1194 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1195 poll_immediate_wake();
1202 ofproto_is_alive(const struct ofproto *p)
1204 return connmgr_has_controllers(p->connmgr);
1207 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1208 * memory_report(). */
1210 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1212 const struct oftable *table;
1213 unsigned int n_rules;
1215 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1216 simap_increase(usage, "ops",
1217 ofproto->n_pending + hmap_count(&ofproto->deletions));
1220 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1221 n_rules += classifier_count(&table->cls);
1223 simap_increase(usage, "rules", n_rules);
1225 if (ofproto->ofproto_class->get_memory_usage) {
1226 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1229 connmgr_get_memory_usage(ofproto->connmgr, usage);
1233 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1236 connmgr_get_controller_info(ofproto->connmgr, info);
1240 ofproto_free_ofproto_controller_info(struct shash *info)
1242 connmgr_free_controller_info(info);
1245 /* Makes a deep copy of 'old' into 'port'. */
1247 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1249 port->name = xstrdup(old->name);
1250 port->type = xstrdup(old->type);
1251 port->ofp_port = old->ofp_port;
1254 /* Frees memory allocated to members of 'ofproto_port'.
1256 * Do not call this function on an ofproto_port obtained from
1257 * ofproto_port_dump_next(): that function retains ownership of the data in the
1260 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1262 free(ofproto_port->name);
1263 free(ofproto_port->type);
1266 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1268 * This function provides no status indication. An error status for the entire
1269 * dump operation is provided when it is completed by calling
1270 * ofproto_port_dump_done().
1273 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1274 const struct ofproto *ofproto)
1276 dump->ofproto = ofproto;
1277 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1281 /* Attempts to retrieve another port from 'dump', which must have been created
1282 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1283 * 'port' and returns true. On failure, returns false.
1285 * Failure might indicate an actual error or merely that the last port has been
1286 * dumped. An error status for the entire dump operation is provided when it
1287 * is completed by calling ofproto_port_dump_done().
1289 * The ofproto owns the data stored in 'port'. It will remain valid until at
1290 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1291 * ofproto_port_dump_done(). */
1293 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1294 struct ofproto_port *port)
1296 const struct ofproto *ofproto = dump->ofproto;
1302 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1305 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1311 /* Completes port table dump operation 'dump', which must have been created
1312 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1313 * error-free, otherwise a positive errno value describing the problem. */
1315 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1317 const struct ofproto *ofproto = dump->ofproto;
1319 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1322 return dump->error == EOF ? 0 : dump->error;
1325 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1326 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1327 * is non-null). On failure, returns a positive errno value and sets
1328 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1330 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1331 uint16_t *ofp_portp)
1336 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1338 update_port(ofproto, netdev_get_name(netdev));
1341 *ofp_portp = error ? OFPP_NONE : ofp_port;
1346 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1347 * initializes '*port' appropriately; on failure, returns a positive errno
1350 * The caller owns the data in 'ofproto_port' and must free it with
1351 * ofproto_port_destroy() when it is no longer needed. */
1353 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1354 struct ofproto_port *port)
1358 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1360 memset(port, 0, sizeof *port);
1365 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1366 * Returns 0 if successful, otherwise a positive errno. */
1368 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1370 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1371 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1374 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1375 if (!error && ofport) {
1376 /* 'name' is the netdev's name and update_port() is going to close the
1377 * netdev. Just in case update_port() refers to 'name' after it
1378 * destroys 'ofport', make a copy of it around the update_port()
1380 char *devname = xstrdup(name);
1381 update_port(ofproto, devname);
1387 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1388 * performs the 'n_actions' actions in 'actions'. The new flow will not
1391 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1392 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1393 * controllers; otherwise, it will be hidden.
1395 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1397 * This is a helper function for in-band control and fail-open. */
1399 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1400 const struct ofpact *ofpacts, size_t ofpacts_len)
1402 const struct rule *rule;
1404 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1405 &ofproto->tables[0].cls, cls_rule));
1406 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1407 ofpacts, ofpacts_len)) {
1408 struct ofputil_flow_mod fm;
1410 memset(&fm, 0, sizeof fm);
1412 fm.buffer_id = UINT32_MAX;
1413 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1414 fm.ofpacts_len = ofpacts_len;
1415 add_flow(ofproto, NULL, &fm, NULL);
1420 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1421 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1422 * operation cannot be initiated now but may be retried later.
1424 * This is a helper function for in-band control and fail-open. */
1426 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1428 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1431 /* Searches for a rule with matching criteria exactly equal to 'target' in
1432 * ofproto's table 0 and, if it finds one, deletes it.
1434 * This is a helper function for in-band control and fail-open. */
1436 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1440 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1441 &ofproto->tables[0].cls, target));
1443 /* No such rule -> success. */
1445 } else if (rule->pending) {
1446 /* An operation on the rule is already pending -> failure.
1447 * Caller must retry later if it's important. */
1450 /* Initiate deletion -> success. */
1451 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1452 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1453 oftable_remove_rule(rule);
1454 ofproto->ofproto_class->rule_destruct(rule);
1455 ofopgroup_submit(group);
1461 /* Starts the process of deleting all of the flows from all of ofproto's flow
1462 * tables and then reintroducing the flows required by in-band control and
1463 * fail-open. The process will complete in a later call to ofproto_run(). */
1465 ofproto_flush_flows(struct ofproto *ofproto)
1467 COVERAGE_INC(ofproto_flush);
1468 ofproto->state = S_FLUSH;
1472 reinit_ports(struct ofproto *p)
1474 struct ofproto_port_dump dump;
1475 struct sset devnames;
1476 struct ofport *ofport;
1477 struct ofproto_port ofproto_port;
1478 const char *devname;
1480 COVERAGE_INC(ofproto_reinit_ports);
1482 sset_init(&devnames);
1483 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1484 sset_add(&devnames, netdev_get_name(ofport->netdev));
1486 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1487 sset_add(&devnames, ofproto_port.name);
1490 SSET_FOR_EACH (devname, &devnames) {
1491 update_port(p, devname);
1493 sset_destroy(&devnames);
1496 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1497 * pointer if the netdev cannot be opened. On success, also fills in
1499 static struct netdev *
1500 ofport_open(const struct ofproto *ofproto,
1501 const struct ofproto_port *ofproto_port,
1502 struct ofputil_phy_port *pp)
1504 enum netdev_flags flags;
1505 struct netdev *netdev;
1508 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1510 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1511 "cannot be opened (%s)",
1513 ofproto_port->name, ofproto_port->ofp_port,
1514 ofproto_port->name, strerror(error));
1518 pp->port_no = ofproto_port->ofp_port;
1519 netdev_get_etheraddr(netdev, pp->hw_addr);
1520 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1521 netdev_get_flags(netdev, &flags);
1522 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1523 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1524 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1525 &pp->supported, &pp->peer);
1526 pp->curr_speed = netdev_features_to_bps(pp->curr);
1527 pp->max_speed = netdev_features_to_bps(pp->supported);
1532 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1533 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1536 ofport_equal(const struct ofputil_phy_port *a,
1537 const struct ofputil_phy_port *b)
1539 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1540 && a->state == b->state
1541 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1542 && a->curr == b->curr
1543 && a->advertised == b->advertised
1544 && a->supported == b->supported
1545 && a->peer == b->peer
1546 && a->curr_speed == b->curr_speed
1547 && a->max_speed == b->max_speed);
1550 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1551 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1552 * one with the same name or port number). */
1554 ofport_install(struct ofproto *p,
1555 struct netdev *netdev, const struct ofputil_phy_port *pp)
1557 const char *netdev_name = netdev_get_name(netdev);
1558 struct ofport *ofport;
1561 /* Create ofport. */
1562 ofport = p->ofproto_class->port_alloc();
1567 ofport->ofproto = p;
1568 ofport->netdev = netdev;
1569 ofport->change_seq = netdev_change_seq(netdev);
1571 ofport->ofp_port = pp->port_no;
1573 /* Add port to 'p'. */
1574 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1575 shash_add(&p->port_by_name, netdev_name, ofport);
1577 update_mtu(p, ofport);
1579 /* Let the ofproto_class initialize its private data. */
1580 error = p->ofproto_class->port_construct(ofport);
1584 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1588 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1589 p->name, netdev_name, strerror(error));
1591 ofport_destroy__(ofport);
1593 netdev_close(netdev);
1597 /* Removes 'ofport' from 'p' and destroys it. */
1599 ofport_remove(struct ofport *ofport)
1601 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1603 ofport_destroy(ofport);
1606 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1609 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1611 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1613 ofport_remove(port);
1617 /* Updates 'port' with new 'pp' description.
1619 * Does not handle a name or port number change. The caller must implement
1620 * such a change as a delete followed by an add. */
1622 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1624 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1625 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1626 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1627 port->pp.state = pp->state;
1628 port->pp.curr = pp->curr;
1629 port->pp.advertised = pp->advertised;
1630 port->pp.supported = pp->supported;
1631 port->pp.peer = pp->peer;
1632 port->pp.curr_speed = pp->curr_speed;
1633 port->pp.max_speed = pp->max_speed;
1635 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1638 /* Update OpenFlow 'state' in 'port' and notify controller. */
1640 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1642 if (port->pp.state != state) {
1643 port->pp.state = state;
1644 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1650 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1652 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1654 if (port->ofproto->ofproto_class->set_realdev) {
1655 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1657 if (port->ofproto->ofproto_class->set_stp_port) {
1658 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1660 if (port->ofproto->ofproto_class->set_cfm) {
1661 port->ofproto->ofproto_class->set_cfm(port, NULL);
1663 if (port->ofproto->ofproto_class->bundle_remove) {
1664 port->ofproto->ofproto_class->bundle_remove(port);
1670 ofport_destroy__(struct ofport *port)
1672 struct ofproto *ofproto = port->ofproto;
1673 const char *name = netdev_get_name(port->netdev);
1675 hmap_remove(&ofproto->ports, &port->hmap_node);
1676 shash_delete(&ofproto->port_by_name,
1677 shash_find(&ofproto->port_by_name, name));
1679 netdev_close(port->netdev);
1680 ofproto->ofproto_class->port_dealloc(port);
1684 ofport_destroy(struct ofport *port)
1687 port->ofproto->ofproto_class->port_destruct(port);
1688 ofport_destroy__(port);
1693 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1695 struct ofport *port;
1697 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1698 hash_int(ofp_port, 0), &ofproto->ports) {
1699 if (port->ofp_port == ofp_port) {
1707 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1709 struct ofproto *ofproto = port->ofproto;
1712 if (ofproto->ofproto_class->port_get_stats) {
1713 error = ofproto->ofproto_class->port_get_stats(port, stats);
1722 update_port(struct ofproto *ofproto, const char *name)
1724 struct ofproto_port ofproto_port;
1725 struct ofputil_phy_port pp;
1726 struct netdev *netdev;
1727 struct ofport *port;
1729 COVERAGE_INC(ofproto_update_port);
1731 /* Fetch 'name''s location and properties from the datapath. */
1732 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1733 ? ofport_open(ofproto, &ofproto_port, &pp)
1736 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1737 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1738 struct netdev *old_netdev = port->netdev;
1740 /* 'name' hasn't changed location. Any properties changed? */
1741 if (!ofport_equal(&port->pp, &pp)) {
1742 ofport_modified(port, &pp);
1745 update_mtu(ofproto, port);
1747 /* Install the newly opened netdev in case it has changed.
1748 * Don't close the old netdev yet in case port_modified has to
1749 * remove a retained reference to it.*/
1750 port->netdev = netdev;
1751 port->change_seq = netdev_change_seq(netdev);
1753 if (port->ofproto->ofproto_class->port_modified) {
1754 port->ofproto->ofproto_class->port_modified(port);
1757 netdev_close(old_netdev);
1759 /* If 'port' is nonnull then its name differs from 'name' and thus
1760 * we should delete it. If we think there's a port named 'name'
1761 * then its port number must be wrong now so delete it too. */
1763 ofport_remove(port);
1765 ofport_remove_with_name(ofproto, name);
1766 ofport_install(ofproto, netdev, &pp);
1769 /* Any port named 'name' is gone now. */
1770 ofport_remove_with_name(ofproto, name);
1772 ofproto_port_destroy(&ofproto_port);
1776 init_ports(struct ofproto *p)
1778 struct ofproto_port_dump dump;
1779 struct ofproto_port ofproto_port;
1781 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1782 uint16_t ofp_port = ofproto_port.ofp_port;
1783 if (ofproto_get_port(p, ofp_port)) {
1784 VLOG_WARN_RL(&rl, "%s: ignoring duplicate port %"PRIu16" "
1785 "in datapath", p->name, ofp_port);
1786 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1787 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1788 p->name, ofproto_port.name);
1790 struct ofputil_phy_port pp;
1791 struct netdev *netdev;
1793 netdev = ofport_open(p, &ofproto_port, &pp);
1795 ofport_install(p, netdev, &pp);
1803 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1804 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1806 find_min_mtu(struct ofproto *p)
1808 struct ofport *ofport;
1811 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1812 struct netdev *netdev = ofport->netdev;
1815 /* Skip any internal ports, since that's what we're trying to
1817 if (!strcmp(netdev_get_type(netdev), "internal")) {
1821 if (netdev_get_mtu(netdev, &dev_mtu)) {
1824 if (!mtu || dev_mtu < mtu) {
1829 return mtu ? mtu: ETH_PAYLOAD_MAX;
1832 /* Update MTU of all datapath devices on 'p' to the minimum of the
1833 * non-datapath ports in event of 'port' added or changed. */
1835 update_mtu(struct ofproto *p, struct ofport *port)
1837 struct ofport *ofport;
1838 struct netdev *netdev = port->netdev;
1839 int dev_mtu, old_min;
1841 if (netdev_get_mtu(netdev, &dev_mtu)) {
1845 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1846 if (dev_mtu > p->min_mtu) {
1847 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1848 dev_mtu = p->min_mtu;
1851 port->mtu = dev_mtu;
1855 /* For non-internal port find new min mtu. */
1856 old_min = p->min_mtu;
1857 port->mtu = dev_mtu;
1858 p->min_mtu = find_min_mtu(p);
1859 if (p->min_mtu == old_min) {
1863 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1864 struct netdev *netdev = ofport->netdev;
1866 if (!strcmp(netdev_get_type(netdev), "internal")) {
1867 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1868 ofport->mtu = p->min_mtu;
1875 ofproto_rule_destroy__(struct rule *rule)
1878 free(rule->ofpacts);
1879 rule->ofproto->ofproto_class->rule_dealloc(rule);
1883 /* This function allows an ofproto implementation to destroy any rules that
1884 * remain when its ->destruct() function is called. The caller must have
1885 * already uninitialized any derived members of 'rule' (step 5 described in the
1886 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1887 * This function implements steps 6 and 7.
1889 * This function should only be called from an ofproto implementation's
1890 * ->destruct() function. It is not suitable elsewhere. */
1892 ofproto_rule_destroy(struct rule *rule)
1894 assert(!rule->pending);
1895 oftable_remove_rule(rule);
1896 ofproto_rule_destroy__(rule);
1899 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1900 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1902 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
1904 return (port == OFPP_NONE
1905 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
1908 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
1909 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
1911 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
1913 if (ofproto_rule_has_out_port(op->rule, out_port)) {
1918 case OFOPERATION_ADD:
1919 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
1921 case OFOPERATION_DELETE:
1924 case OFOPERATION_MODIFY:
1925 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
1931 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1932 * statistics appropriately. 'packet' must have at least sizeof(struct
1933 * ofp_packet_in) bytes of headroom.
1935 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1936 * with statistics for 'packet' either way.
1938 * Takes ownership of 'packet'. */
1940 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1944 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1946 flow_extract(packet, 0, 0, in_port, &flow);
1947 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1950 /* Returns true if 'rule' should be hidden from the controller.
1952 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1953 * (e.g. by in-band control) and are intentionally hidden from the
1956 ofproto_rule_is_hidden(const struct rule *rule)
1958 return rule->cr.priority > UINT16_MAX;
1961 static enum oftable_flags
1962 rule_get_flags(const struct rule *rule)
1964 return rule->ofproto->tables[rule->table_id].flags;
1968 rule_is_modifiable(const struct rule *rule)
1970 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1974 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1976 ofconn_send_reply(ofconn, make_echo_reply(oh));
1981 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1983 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1984 struct ofputil_switch_features features;
1985 struct ofport *port;
1989 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1991 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1993 features.datapath_id = ofproto->datapath_id;
1994 features.n_buffers = pktbuf_capacity();
1995 features.n_tables = ofproto->n_tables;
1996 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1997 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1999 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2002 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2004 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2005 ofputil_put_switch_features_port(&port->pp, b);
2008 ofconn_send_reply(ofconn, b);
2013 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2015 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2016 struct ofp_switch_config *osc;
2017 enum ofp_config_flags flags;
2021 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
2022 flags = ofproto->frag_handling;
2023 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
2024 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2026 osc->flags = htons(flags);
2027 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2028 ofconn_send_reply(ofconn, buf);
2034 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2036 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2037 uint16_t flags = ntohs(osc->flags);
2039 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2040 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2041 enum ofp_config_flags cur = ofproto->frag_handling;
2042 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2044 assert((cur & OFPC_FRAG_MASK) == cur);
2046 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2047 ofproto->frag_handling = next;
2049 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2051 ofputil_frag_handling_to_string(next));
2055 ofconn_set_invalid_ttl_to_controller(ofconn,
2056 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2058 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2063 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2064 * error message code for the caller to propagate upward. Otherwise, returns
2067 * The log message mentions 'msg_type'. */
2069 reject_slave_controller(struct ofconn *ofconn)
2071 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2072 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2073 return OFPERR_OFPBRC_EPERM;
2080 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
2082 struct ofproto *p = ofconn_get_ofproto(ofconn);
2083 struct ofputil_packet_out po;
2084 struct ofpbuf *payload;
2085 uint64_t ofpacts_stub[1024 / 8];
2086 struct ofpbuf ofpacts;
2090 COVERAGE_INC(ofproto_packet_out);
2092 error = reject_slave_controller(ofconn);
2097 /* Decode message. */
2098 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2099 error = ofputil_decode_packet_out(&po, opo, &ofpacts);
2101 goto exit_free_ofpacts;
2105 if (po.buffer_id != UINT32_MAX) {
2106 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2107 if (error || !payload) {
2108 goto exit_free_ofpacts;
2111 payload = xmalloc(sizeof *payload);
2112 ofpbuf_use_const(payload, po.packet, po.packet_len);
2115 /* Send out packet. */
2116 flow_extract(payload, 0, 0, po.in_port, &flow);
2117 error = p->ofproto_class->packet_out(p, payload, &flow,
2118 po.ofpacts, po.ofpacts_len);
2119 ofpbuf_delete(payload);
2122 ofpbuf_uninit(&ofpacts);
2128 update_port_config(struct ofport *port,
2129 enum ofputil_port_config config,
2130 enum ofputil_port_config mask)
2132 enum ofputil_port_config old_config = port->pp.config;
2133 enum ofputil_port_config toggle;
2135 toggle = (config ^ port->pp.config) & mask;
2136 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2137 if (config & OFPUTIL_PC_PORT_DOWN) {
2138 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2140 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2142 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2145 port->pp.config ^= toggle;
2146 if (port->pp.config != old_config) {
2147 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2152 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2154 struct ofproto *p = ofconn_get_ofproto(ofconn);
2155 struct ofputil_port_mod pm;
2156 struct ofport *port;
2159 error = reject_slave_controller(ofconn);
2164 error = ofputil_decode_port_mod(oh, &pm);
2169 port = ofproto_get_port(p, pm.port_no);
2171 return OFPERR_OFPPMFC_BAD_PORT;
2172 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2173 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2175 update_port_config(port, pm.config, pm.mask);
2177 netdev_set_advertisements(port->netdev, pm.advertise);
2184 handle_desc_stats_request(struct ofconn *ofconn,
2185 const struct ofp_stats_msg *request)
2187 struct ofproto *p = ofconn_get_ofproto(ofconn);
2188 struct ofp_desc_stats *ods;
2191 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2192 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2193 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2194 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2195 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2196 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2197 ofconn_send_reply(ofconn, msg);
2203 handle_table_stats_request(struct ofconn *ofconn,
2204 const struct ofp_stats_msg *request)
2206 struct ofproto *p = ofconn_get_ofproto(ofconn);
2207 struct ofp_table_stats *ots;
2211 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2213 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2214 for (i = 0; i < p->n_tables; i++) {
2215 ots[i].table_id = i;
2216 sprintf(ots[i].name, "table%zu", i);
2217 ots[i].wildcards = htonl(OFPFW10_ALL);
2218 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2219 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2222 p->ofproto_class->get_tables(p, ots);
2224 for (i = 0; i < p->n_tables; i++) {
2225 const struct oftable *table = &p->tables[i];
2228 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2231 if (table->max_flows < ntohl(ots[i].max_entries)) {
2232 ots[i].max_entries = htonl(table->max_flows);
2236 ofconn_send_reply(ofconn, msg);
2241 append_port_stat(struct ofport *port, struct list *replies)
2243 struct netdev_stats stats;
2244 struct ofp_port_stats *ops;
2246 /* Intentionally ignore return value, since errors will set
2247 * 'stats' to all-1s, which is correct for OpenFlow, and
2248 * netdev_get_stats() will log errors. */
2249 ofproto_port_get_stats(port, &stats);
2251 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2252 ops->port_no = htons(port->pp.port_no);
2253 memset(ops->pad, 0, sizeof ops->pad);
2254 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2255 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2256 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2257 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2258 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2259 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2260 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2261 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2262 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2263 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2264 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2265 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2269 handle_port_stats_request(struct ofconn *ofconn,
2270 const struct ofp_port_stats_request *psr)
2272 struct ofproto *p = ofconn_get_ofproto(ofconn);
2273 struct ofport *port;
2274 struct list replies;
2276 ofputil_start_stats_reply(&psr->osm, &replies);
2277 if (psr->port_no != htons(OFPP_NONE)) {
2278 port = ofproto_get_port(p, ntohs(psr->port_no));
2280 append_port_stat(port, &replies);
2283 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2284 append_port_stat(port, &replies);
2288 ofconn_send_replies(ofconn, &replies);
2293 handle_port_desc_stats_request(struct ofconn *ofconn,
2294 const struct ofp_stats_msg *osm)
2296 struct ofproto *p = ofconn_get_ofproto(ofconn);
2297 struct ofport *port;
2298 struct list replies;
2300 ofputil_start_stats_reply(osm, &replies);
2302 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2303 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2304 &port->pp, &replies);
2307 ofconn_send_replies(ofconn, &replies);
2312 calc_flow_duration__(long long int start, long long int now,
2313 uint32_t *sec, uint32_t *nsec)
2315 long long int msecs = now - start;
2316 *sec = msecs / 1000;
2317 *nsec = (msecs % 1000) * (1000 * 1000);
2320 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2321 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2323 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2325 return (table_id == 0xff || table_id < ofproto->n_tables
2327 : OFPERR_NXBRC_BAD_TABLE_ID);
2331 static struct oftable *
2332 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2334 struct oftable *table;
2336 for (table = &ofproto->tables[table_id];
2337 table < &ofproto->tables[ofproto->n_tables];
2339 if (!(table->flags & OFTABLE_HIDDEN)) {
2347 static struct oftable *
2348 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2350 if (table_id == 0xff) {
2351 return next_visible_table(ofproto, 0);
2352 } else if (table_id < ofproto->n_tables) {
2353 return &ofproto->tables[table_id];
2359 static struct oftable *
2360 next_matching_table(const struct ofproto *ofproto,
2361 const struct oftable *table, uint8_t table_id)
2363 return (table_id == 0xff
2364 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2368 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2370 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2371 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2373 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2374 * only once, for that table. (This can be used to access tables marked
2377 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2378 * entered at all. (Perhaps you should have validated TABLE_ID with
2379 * check_table_id().)
2381 * All parameters are evaluated multiple times.
2383 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2384 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2386 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2388 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2389 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2390 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2393 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2394 * to 'out_port' are included.
2396 * Hidden rules are always omitted.
2398 * Returns 0 on success, otherwise an OpenFlow error code. */
2400 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2401 const struct cls_rule *match,
2402 ovs_be64 cookie, ovs_be64 cookie_mask,
2403 uint16_t out_port, struct list *rules)
2405 struct oftable *table;
2408 error = check_table_id(ofproto, table_id);
2414 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2415 struct cls_cursor cursor;
2418 cls_cursor_init(&cursor, &table->cls, match);
2419 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2420 if (rule->pending) {
2421 return OFPROTO_POSTPONE;
2423 if (!ofproto_rule_is_hidden(rule)
2424 && ofproto_rule_has_out_port(rule, out_port)
2425 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2426 list_push_back(rules, &rule->ofproto_node);
2433 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2434 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2435 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2438 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2439 * to 'out_port' are included.
2441 * Hidden rules are always omitted.
2443 * Returns 0 on success, otherwise an OpenFlow error code. */
2445 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2446 const struct cls_rule *match,
2447 ovs_be64 cookie, ovs_be64 cookie_mask,
2448 uint16_t out_port, struct list *rules)
2450 struct oftable *table;
2453 error = check_table_id(ofproto, table_id);
2459 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2462 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2465 if (rule->pending) {
2466 return OFPROTO_POSTPONE;
2468 if (!ofproto_rule_is_hidden(rule)
2469 && ofproto_rule_has_out_port(rule, out_port)
2470 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2471 list_push_back(rules, &rule->ofproto_node);
2478 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2479 * forced into the range of a uint16_t. */
2481 age_secs(long long int age_ms)
2483 return (age_ms < 0 ? 0
2484 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2485 : (unsigned int) age_ms / 1000);
2489 handle_flow_stats_request(struct ofconn *ofconn,
2490 const struct ofp_stats_msg *osm)
2492 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2493 struct ofputil_flow_stats_request fsr;
2494 struct list replies;
2499 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2504 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2505 fsr.cookie, fsr.cookie_mask,
2506 fsr.out_port, &rules);
2511 ofputil_start_stats_reply(osm, &replies);
2512 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2513 long long int now = time_msec();
2514 struct ofputil_flow_stats fs;
2517 fs.cookie = rule->flow_cookie;
2518 fs.table_id = rule->table_id;
2519 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2521 fs.idle_timeout = rule->idle_timeout;
2522 fs.hard_timeout = rule->hard_timeout;
2523 fs.idle_age = age_secs(now - rule->used);
2524 fs.hard_age = age_secs(now - rule->modified);
2525 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2527 fs.ofpacts = rule->ofpacts;
2528 fs.ofpacts_len = rule->ofpacts_len;
2529 ofputil_append_flow_stats_reply(&fs, &replies);
2531 ofconn_send_replies(ofconn, &replies);
2537 flow_stats_ds(struct rule *rule, struct ds *results)
2539 uint64_t packet_count, byte_count;
2541 rule->ofproto->ofproto_class->rule_get_stats(rule,
2542 &packet_count, &byte_count);
2544 if (rule->table_id != 0) {
2545 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2547 ds_put_format(results, "duration=%llds, ",
2548 (time_msec() - rule->created) / 1000);
2549 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2550 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2551 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2552 cls_rule_format(&rule->cr, results);
2553 ds_put_char(results, ',');
2554 if (rule->ofpacts_len > 0) {
2555 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2557 ds_put_cstr(results, "drop");
2559 ds_put_cstr(results, "\n");
2562 /* Adds a pretty-printed description of all flows to 'results', including
2563 * hidden flows (e.g., set up by in-band control). */
2565 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2567 struct oftable *table;
2569 OFPROTO_FOR_EACH_TABLE (table, p) {
2570 struct cls_cursor cursor;
2573 cls_cursor_init(&cursor, &table->cls, NULL);
2574 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2575 flow_stats_ds(rule, results);
2580 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2581 * '*engine_type' and '*engine_id', respectively. */
2583 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2584 uint8_t *engine_type, uint8_t *engine_id)
2586 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2589 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2590 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2591 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2592 * and -1 if CFM is not enabled on 'port'. */
2594 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2596 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2597 return (ofport && ofproto->ofproto_class->get_cfm_fault
2598 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2602 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2603 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2604 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2605 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2607 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2608 uint16_t ofp_port, const uint64_t **rmps,
2611 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2615 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2616 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2621 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2622 * integer value between 0 and 100 to indicate the health of the port as a
2623 * percentage which is the average of cfm health of all the remote_mpids or
2624 * returns -1 if CFM is not enabled on 'ofport'. */
2626 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2628 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2629 return (ofport && ofproto->ofproto_class->get_cfm_health
2630 ? ofproto->ofproto_class->get_cfm_health(ofport)
2635 handle_aggregate_stats_request(struct ofconn *ofconn,
2636 const struct ofp_stats_msg *osm)
2638 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2639 struct ofputil_flow_stats_request request;
2640 struct ofputil_aggregate_stats stats;
2641 bool unknown_packets, unknown_bytes;
2642 struct ofpbuf *reply;
2647 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2652 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2653 request.cookie, request.cookie_mask,
2654 request.out_port, &rules);
2659 memset(&stats, 0, sizeof stats);
2660 unknown_packets = unknown_bytes = false;
2661 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2662 uint64_t packet_count;
2663 uint64_t byte_count;
2665 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2668 if (packet_count == UINT64_MAX) {
2669 unknown_packets = true;
2671 stats.packet_count += packet_count;
2674 if (byte_count == UINT64_MAX) {
2675 unknown_bytes = true;
2677 stats.byte_count += byte_count;
2682 if (unknown_packets) {
2683 stats.packet_count = UINT64_MAX;
2685 if (unknown_bytes) {
2686 stats.byte_count = UINT64_MAX;
2689 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2690 ofconn_send_reply(ofconn, reply);
2695 struct queue_stats_cbdata {
2696 struct ofport *ofport;
2697 struct list replies;
2701 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2702 const struct netdev_queue_stats *stats)
2704 struct ofp_queue_stats *reply;
2706 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2707 reply->port_no = htons(cbdata->ofport->pp.port_no);
2708 memset(reply->pad, 0, sizeof reply->pad);
2709 reply->queue_id = htonl(queue_id);
2710 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2711 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2712 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2716 handle_queue_stats_dump_cb(uint32_t queue_id,
2717 struct netdev_queue_stats *stats,
2720 struct queue_stats_cbdata *cbdata = cbdata_;
2722 put_queue_stats(cbdata, queue_id, stats);
2726 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2727 struct queue_stats_cbdata *cbdata)
2729 cbdata->ofport = port;
2730 if (queue_id == OFPQ_ALL) {
2731 netdev_dump_queue_stats(port->netdev,
2732 handle_queue_stats_dump_cb, cbdata);
2734 struct netdev_queue_stats stats;
2736 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2737 put_queue_stats(cbdata, queue_id, &stats);
2739 return OFPERR_OFPQOFC_BAD_QUEUE;
2746 handle_queue_stats_request(struct ofconn *ofconn,
2747 const struct ofp_queue_stats_request *qsr)
2749 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2750 struct queue_stats_cbdata cbdata;
2751 unsigned int port_no;
2752 struct ofport *port;
2756 COVERAGE_INC(ofproto_queue_req);
2758 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2760 port_no = ntohs(qsr->port_no);
2761 queue_id = ntohl(qsr->queue_id);
2762 if (port_no == OFPP_ALL) {
2763 error = OFPERR_OFPQOFC_BAD_QUEUE;
2764 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2765 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2770 port = ofproto_get_port(ofproto, port_no);
2772 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2773 : OFPERR_OFPQOFC_BAD_PORT);
2776 ofconn_send_replies(ofconn, &cbdata.replies);
2778 ofpbuf_list_delete(&cbdata.replies);
2785 is_flow_deletion_pending(const struct ofproto *ofproto,
2786 const struct cls_rule *cls_rule,
2789 if (!hmap_is_empty(&ofproto->deletions)) {
2790 struct ofoperation *op;
2792 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2793 cls_rule_hash(cls_rule, table_id),
2794 &ofproto->deletions) {
2795 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2804 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2805 * in which no matching flow already exists in the flow table.
2807 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2808 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2809 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2810 * initiated now but may be retried later.
2812 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2813 * ownership remains with the caller.
2815 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2818 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2819 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2821 struct oftable *table;
2822 struct ofopgroup *group;
2823 struct rule *victim;
2827 error = check_table_id(ofproto, fm->table_id);
2833 if (fm->table_id == 0xff) {
2835 if (ofproto->ofproto_class->rule_choose_table) {
2836 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2841 assert(table_id < ofproto->n_tables);
2842 table = &ofproto->tables[table_id];
2844 table = &ofproto->tables[0];
2846 } else if (fm->table_id < ofproto->n_tables) {
2847 table = &ofproto->tables[fm->table_id];
2849 return OFPERR_NXFMFC_BAD_TABLE_ID;
2852 if (table->flags & OFTABLE_READONLY) {
2853 return OFPERR_OFPBRC_EPERM;
2856 /* Check for overlap, if requested. */
2857 if (fm->flags & OFPFF_CHECK_OVERLAP
2858 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2859 return OFPERR_OFPFMFC_OVERLAP;
2862 /* Serialize against pending deletion. */
2863 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2864 return OFPROTO_POSTPONE;
2867 /* Allocate new rule. */
2868 rule = ofproto->ofproto_class->rule_alloc();
2870 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2871 ofproto->name, strerror(error));
2874 rule->ofproto = ofproto;
2876 rule->pending = NULL;
2877 rule->flow_cookie = fm->new_cookie;
2878 rule->created = rule->modified = rule->used = time_msec();
2879 rule->idle_timeout = fm->idle_timeout;
2880 rule->hard_timeout = fm->hard_timeout;
2881 rule->table_id = table - ofproto->tables;
2882 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2883 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2884 rule->ofpacts_len = fm->ofpacts_len;
2885 rule->evictable = true;
2886 rule->eviction_group = NULL;
2887 rule->monitor_flags = 0;
2888 rule->add_seqno = 0;
2889 rule->modify_seqno = 0;
2891 /* Insert new rule. */
2892 victim = oftable_replace_rule(rule);
2893 if (victim && !rule_is_modifiable(victim)) {
2894 error = OFPERR_OFPBRC_EPERM;
2895 } else if (victim && victim->pending) {
2896 error = OFPROTO_POSTPONE;
2898 struct ofoperation *op;
2901 if (classifier_count(&table->cls) > table->max_flows) {
2904 was_evictable = rule->evictable;
2905 rule->evictable = false;
2906 evict = choose_rule_to_evict(table);
2907 rule->evictable = was_evictable;
2910 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2912 } else if (evict->pending) {
2913 error = OFPROTO_POSTPONE;
2920 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2921 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
2922 op->victim = victim;
2924 error = ofproto->ofproto_class->rule_construct(rule);
2926 op->group->n_running--;
2927 ofoperation_destroy(rule->pending);
2929 delete_flow__(evict, group);
2931 ofopgroup_submit(group);
2935 /* Back out if an error occurred. */
2937 oftable_substitute_rule(rule, victim);
2938 ofproto_rule_destroy__(rule);
2943 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2945 /* Modifies the rules listed in 'rules', changing their actions to match those
2948 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2951 * Returns 0 on success, otherwise an OpenFlow error code. */
2953 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2954 const struct ofputil_flow_mod *fm,
2955 const struct ofp_header *request, struct list *rules)
2957 struct ofopgroup *group;
2961 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2962 error = OFPERR_OFPBRC_EPERM;
2963 LIST_FOR_EACH (rule, ofproto_node, rules) {
2964 struct ofoperation *op;
2965 bool actions_changed;
2966 ovs_be64 new_cookie;
2968 if (rule_is_modifiable(rule)) {
2969 /* At least one rule is modifiable, don't report EPERM error. */
2975 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2976 rule->ofpacts, rule->ofpacts_len);
2977 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
2979 : rule->flow_cookie);
2980 if (!actions_changed && new_cookie == rule->flow_cookie) {
2981 /* No change at all. */
2985 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
2986 rule->flow_cookie = new_cookie;
2987 if (actions_changed) {
2988 op->ofpacts = rule->ofpacts;
2989 op->ofpacts_len = rule->ofpacts_len;
2990 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2991 rule->ofpacts_len = fm->ofpacts_len;
2992 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2994 ofoperation_complete(op, 0);
2997 ofopgroup_submit(group);
3002 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3005 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3008 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3009 const struct ofputil_flow_mod *fm,
3010 const struct ofp_header *request)
3015 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3016 fm->cookie, fm->cookie_mask,
3020 } else if (list_is_empty(&rules)) {
3021 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
3023 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3027 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3030 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3033 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3034 const struct ofputil_flow_mod *fm,
3035 const struct ofp_header *request)
3040 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3041 fm->cookie, fm->cookie_mask,
3046 } else if (list_is_empty(&rules)) {
3047 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
3049 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3050 fm, request, &rules)
3055 /* OFPFC_DELETE implementation. */
3058 delete_flow__(struct rule *rule, struct ofopgroup *group)
3060 struct ofproto *ofproto = rule->ofproto;
3062 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3064 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3065 oftable_remove_rule(rule);
3066 ofproto->ofproto_class->rule_destruct(rule);
3069 /* Deletes the rules listed in 'rules'.
3071 * Returns 0 on success, otherwise an OpenFlow error code. */
3073 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3074 const struct ofp_header *request, struct list *rules)
3076 struct rule *rule, *next;
3077 struct ofopgroup *group;
3079 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3080 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3081 delete_flow__(rule, group);
3083 ofopgroup_submit(group);
3088 /* Implements OFPFC_DELETE. */
3090 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3091 const struct ofputil_flow_mod *fm,
3092 const struct ofp_header *request)
3097 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3098 fm->cookie, fm->cookie_mask,
3099 fm->out_port, &rules);
3100 return (error ? error
3101 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3106 /* Implements OFPFC_DELETE_STRICT. */
3108 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3109 const struct ofputil_flow_mod *fm,
3110 const struct ofp_header *request)
3115 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3116 fm->cookie, fm->cookie_mask,
3117 fm->out_port, &rules);
3118 return (error ? error
3119 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3125 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3127 struct ofputil_flow_removed fr;
3129 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3134 fr.cookie = rule->flow_cookie;
3136 calc_flow_duration__(rule->created, time_msec(),
3137 &fr.duration_sec, &fr.duration_nsec);
3138 fr.idle_timeout = rule->idle_timeout;
3139 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3142 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3146 ofproto_rule_update_used(struct rule *rule, long long int used)
3148 if (used > rule->used) {
3149 struct eviction_group *evg = rule->eviction_group;
3153 heap_change(&evg->rules, &rule->evg_node,
3154 rule_eviction_priority(rule));
3159 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3160 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3163 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3166 * ofproto implementation ->run() functions should use this function to expire
3167 * OpenFlow flows. */
3169 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3171 struct ofproto *ofproto = rule->ofproto;
3172 struct ofopgroup *group;
3174 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3176 ofproto_rule_send_removed(rule, reason);
3178 group = ofopgroup_create_unattached(ofproto);
3179 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3180 oftable_remove_rule(rule);
3181 ofproto->ofproto_class->rule_destruct(rule);
3182 ofopgroup_submit(group);
3186 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3188 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3189 struct ofputil_flow_mod fm;
3190 uint64_t ofpacts_stub[1024 / 8];
3191 struct ofpbuf ofpacts;
3195 error = reject_slave_controller(ofconn);
3200 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3201 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3204 goto exit_free_ofpacts;
3207 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3208 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3209 if (fm.flags & OFPFF_EMERG) {
3210 /* We do not support the emergency flow cache. It will hopefully get
3211 * dropped from OpenFlow in the near future. There is no good error
3212 * code, so just state that the flow table is full. */
3213 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
3215 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3218 goto exit_free_ofpacts;
3221 /* Record the operation for logging a summary report. */
3222 switch (fm.command) {
3228 case OFPFC_MODIFY_STRICT:
3229 ofproto->n_modify++;
3233 case OFPFC_DELETE_STRICT:
3234 ofproto->n_delete++;
3239 if (ofproto->next_op_report == LLONG_MAX) {
3240 ofproto->first_op = now;
3241 ofproto->next_op_report = MAX(now + 10 * 1000,
3242 ofproto->op_backoff);
3243 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3245 ofproto->last_op = now;
3248 ofpbuf_uninit(&ofpacts);
3254 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3255 const struct ofputil_flow_mod *fm,
3256 const struct ofp_header *oh)
3258 if (ofproto->n_pending >= 50) {
3259 assert(!list_is_empty(&ofproto->pending));
3260 return OFPROTO_POSTPONE;
3263 switch (fm->command) {
3265 return add_flow(ofproto, ofconn, fm, oh);
3268 return modify_flows_loose(ofproto, ofconn, fm, oh);
3270 case OFPFC_MODIFY_STRICT:
3271 return modify_flow_strict(ofproto, ofconn, fm, oh);
3274 return delete_flows_loose(ofproto, ofconn, fm, oh);
3276 case OFPFC_DELETE_STRICT:
3277 return delete_flow_strict(ofproto, ofconn, fm, oh);
3280 if (fm->command > 0xff) {
3281 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3282 "flow_mod_table_id extension is not enabled",
3285 return OFPERR_OFPFMFC_BAD_COMMAND;
3290 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3292 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3293 struct nx_role_request *reply;
3297 role = ntohl(nrr->role);
3298 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3299 && role != NX_ROLE_SLAVE) {
3300 return OFPERR_OFPRRFC_BAD_ROLE;
3303 if (ofconn_get_role(ofconn) != role
3304 && ofconn_has_pending_opgroups(ofconn)) {
3305 return OFPROTO_POSTPONE;
3308 ofconn_set_role(ofconn, role);
3310 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3311 reply->role = htonl(role);
3312 ofconn_send_reply(ofconn, buf);
3318 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3319 const struct ofp_header *oh)
3321 const struct nx_flow_mod_table_id *msg
3322 = (const struct nx_flow_mod_table_id *) oh;
3323 enum ofputil_protocol cur, next;
3325 cur = ofconn_get_protocol(ofconn);
3326 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3327 ofconn_set_protocol(ofconn, next);
3333 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3335 const struct nx_set_flow_format *msg
3336 = (const struct nx_set_flow_format *) oh;
3337 enum ofputil_protocol cur, next;
3338 enum ofputil_protocol next_base;
3340 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3342 return OFPERR_OFPBRC_EPERM;
3345 cur = ofconn_get_protocol(ofconn);
3346 next = ofputil_protocol_set_base(cur, next_base);
3347 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3348 /* Avoid sending async messages in surprising protocol. */
3349 return OFPROTO_POSTPONE;
3352 ofconn_set_protocol(ofconn, next);
3357 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3358 const struct ofp_header *oh)
3360 const struct nx_set_packet_in_format *msg;
3363 msg = (const struct nx_set_packet_in_format *) oh;
3364 format = ntohl(msg->format);
3365 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3366 return OFPERR_OFPBRC_EPERM;
3369 if (format != ofconn_get_packet_in_format(ofconn)
3370 && ofconn_has_pending_opgroups(ofconn)) {
3371 /* Avoid sending async message in surprsing packet in format. */
3372 return OFPROTO_POSTPONE;
3375 ofconn_set_packet_in_format(ofconn, format);
3380 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3382 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3383 uint32_t master[OAM_N_TYPES];
3384 uint32_t slave[OAM_N_TYPES];
3386 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3387 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3388 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3390 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3391 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3392 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3394 ofconn_set_async_config(ofconn, master, slave);
3395 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3396 !ofconn_get_miss_send_len(ofconn)) {
3397 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3404 handle_nxt_set_controller_id(struct ofconn *ofconn,
3405 const struct ofp_header *oh)
3407 const struct nx_controller_id *nci;
3409 nci = (const struct nx_controller_id *) oh;
3410 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3411 return OFPERR_NXBRC_MUST_BE_ZERO;
3414 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3419 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3423 if (ofconn_has_pending_opgroups(ofconn)) {
3424 return OFPROTO_POSTPONE;
3427 make_openflow_xid(sizeof *oh, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3428 ofconn_send_reply(ofconn, buf);
3433 ofproto_compose_flow_refresh_update(const struct rule *rule,
3434 enum nx_flow_monitor_flags flags,
3437 struct ofoperation *op = rule->pending;
3438 struct ofputil_flow_update fu;
3440 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3441 /* We'll report the final flow when the operation completes. Reporting
3442 * it now would cause a duplicate report later. */
3446 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3447 ? NXFME_ADDED : NXFME_MODIFIED);
3449 fu.idle_timeout = rule->idle_timeout;
3450 fu.hard_timeout = rule->hard_timeout;
3451 fu.table_id = rule->table_id;
3452 fu.cookie = rule->flow_cookie;
3453 fu.match = (struct cls_rule *) &rule->cr;
3454 if (!(flags & NXFMF_ACTIONS)) {
3458 fu.ofpacts = rule->ofpacts;
3459 fu.ofpacts_len = rule->ofpacts_len;
3461 /* An operation is in progress. Use the previous version of the flow's
3462 * actions, so that when the operation commits we report the change. */
3464 case OFOPERATION_ADD:
3465 /* We already verified that there was a victim. */
3466 fu.ofpacts = op->victim->ofpacts;
3467 fu.ofpacts_len = op->victim->ofpacts_len;
3470 case OFOPERATION_MODIFY:
3472 fu.ofpacts = op->ofpacts;
3473 fu.ofpacts_len = op->ofpacts_len;
3475 fu.ofpacts = rule->ofpacts;
3476 fu.ofpacts_len = rule->ofpacts_len;
3480 case OFOPERATION_DELETE:
3481 fu.ofpacts = rule->ofpacts;
3482 fu.ofpacts_len = rule->ofpacts_len;
3490 if (list_is_empty(msgs)) {
3491 ofputil_start_flow_update(msgs);
3493 ofputil_append_flow_update(&fu, msgs);
3497 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3501 LIST_FOR_EACH (rule, ofproto_node, rules) {
3502 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3503 rule->monitor_flags = 0;
3505 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3510 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3511 struct rule *rule, uint64_t seqno,
3514 enum nx_flow_monitor_flags update;
3516 if (ofproto_rule_is_hidden(rule)) {
3521 ? ofoperation_has_out_port(rule->pending, m->out_port)
3522 : ofproto_rule_has_out_port(rule, m->out_port))) {
3527 if (rule->add_seqno > seqno) {
3528 update = NXFMF_ADD | NXFMF_MODIFY;
3529 } else if (rule->modify_seqno > seqno) {
3530 update = NXFMF_MODIFY;
3535 if (!(m->flags & update)) {
3539 update = NXFMF_INITIAL;
3542 if (!rule->monitor_flags) {
3543 list_push_back(rules, &rule->ofproto_node);
3545 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3549 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3553 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3554 const struct ofoperation *op;
3555 const struct oftable *table;
3557 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3558 struct cls_cursor cursor;
3561 cls_cursor_init(&cursor, &table->cls, &m->match);
3562 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3563 assert(!rule->pending); /* XXX */
3564 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3568 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3569 struct rule *rule = op->rule;
3571 if (((m->table_id == 0xff
3572 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3573 : m->table_id == rule->table_id))
3574 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
3575 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3581 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3584 if (m->flags & NXFMF_INITIAL) {
3585 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3590 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3591 uint64_t seqno, struct list *rules)
3593 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3597 handle_flow_monitor_request(struct ofconn *ofconn,
3598 const struct ofp_stats_msg *osm)
3600 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3601 struct ofmonitor **monitors;
3602 size_t n_monitors, allocated_monitors;
3603 struct list replies;
3610 ofpbuf_use_const(&b, osm, ntohs(osm->header.length));
3612 n_monitors = allocated_monitors = 0;
3614 struct ofputil_flow_monitor_request request;
3615 struct ofmonitor *m;
3618 retval = ofputil_decode_flow_monitor_request(&request, &b);
3619 if (retval == EOF) {
3621 } else if (retval) {
3626 if (request.table_id != 0xff
3627 && request.table_id >= ofproto->n_tables) {
3628 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3632 error = ofmonitor_create(&request, ofconn, &m);
3637 if (n_monitors >= allocated_monitors) {
3638 monitors = x2nrealloc(monitors, &allocated_monitors,
3641 monitors[n_monitors++] = m;
3645 for (i = 0; i < n_monitors; i++) {
3646 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3649 ofputil_start_stats_reply(osm, &replies);
3650 ofmonitor_compose_refresh_updates(&rules, &replies);
3651 ofconn_send_replies(ofconn, &replies);
3658 for (i = 0; i < n_monitors; i++) {
3659 ofmonitor_destroy(monitors[i]);
3666 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3668 struct ofmonitor *m;
3671 id = ofputil_decode_flow_monitor_cancel(oh);
3672 m = ofmonitor_lookup(ofconn, id);
3674 return OFPERR_NXBRC_FM_BAD_ID;
3677 ofmonitor_destroy(m);
3682 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3684 const struct ofp_header *oh = msg->data;
3685 const struct ofputil_msg_type *type;
3688 error = ofputil_decode_msg_type(oh, &type);
3693 switch (ofputil_msg_type_code(type)) {
3694 /* OpenFlow requests. */
3695 case OFPUTIL_OFPT_ECHO_REQUEST:
3696 return handle_echo_request(ofconn, oh);
3698 case OFPUTIL_OFPT_FEATURES_REQUEST:
3699 return handle_features_request(ofconn, oh);
3701 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3702 return handle_get_config_request(ofconn, oh);
3704 case OFPUTIL_OFPT_SET_CONFIG:
3705 return handle_set_config(ofconn, msg->data);
3707 case OFPUTIL_OFPT_PACKET_OUT:
3708 return handle_packet_out(ofconn, msg->data);
3710 case OFPUTIL_OFPT_PORT_MOD:
3711 return handle_port_mod(ofconn, oh);
3713 case OFPUTIL_OFPT_FLOW_MOD:
3714 return handle_flow_mod(ofconn, oh);
3716 case OFPUTIL_OFPT_BARRIER_REQUEST:
3717 return handle_barrier_request(ofconn, oh);
3719 /* OpenFlow replies. */
3720 case OFPUTIL_OFPT_ECHO_REPLY:
3723 /* Nicira extension requests. */
3724 case OFPUTIL_NXT_ROLE_REQUEST:
3725 return handle_role_request(ofconn, oh);
3727 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3728 return handle_nxt_flow_mod_table_id(ofconn, oh);
3730 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3731 return handle_nxt_set_flow_format(ofconn, oh);
3733 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3734 return handle_nxt_set_packet_in_format(ofconn, oh);
3736 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3737 return handle_nxt_set_controller_id(ofconn, oh);
3739 case OFPUTIL_NXT_FLOW_MOD:
3740 return handle_flow_mod(ofconn, oh);
3742 case OFPUTIL_NXT_FLOW_AGE:
3743 /* Nothing to do. */
3746 case OFPUTIL_NXT_FLOW_MONITOR_CANCEL:
3747 return handle_flow_monitor_cancel(ofconn, oh);
3749 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3750 return handle_nxt_set_async_config(ofconn, oh);
3752 /* Statistics requests. */
3753 case OFPUTIL_OFPST_DESC_REQUEST:
3754 return handle_desc_stats_request(ofconn, msg->data);
3756 case OFPUTIL_OFPST_FLOW_REQUEST:
3757 case OFPUTIL_NXST_FLOW_REQUEST:
3758 return handle_flow_stats_request(ofconn, msg->data);
3760 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3761 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3762 return handle_aggregate_stats_request(ofconn, msg->data);
3764 case OFPUTIL_OFPST_TABLE_REQUEST:
3765 return handle_table_stats_request(ofconn, msg->data);
3767 case OFPUTIL_OFPST_PORT_REQUEST:
3768 return handle_port_stats_request(ofconn, msg->data);
3770 case OFPUTIL_OFPST_QUEUE_REQUEST:
3771 return handle_queue_stats_request(ofconn, msg->data);
3773 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3774 return handle_port_desc_stats_request(ofconn, msg->data);
3776 case OFPUTIL_NXST_FLOW_MONITOR_REQUEST:
3777 return handle_flow_monitor_request(ofconn, msg->data);
3779 case OFPUTIL_MSG_INVALID:
3780 case OFPUTIL_OFPT_HELLO:
3781 case OFPUTIL_OFPT_ERROR:
3782 case OFPUTIL_OFPT_FEATURES_REPLY:
3783 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3784 case OFPUTIL_OFPT_PACKET_IN:
3785 case OFPUTIL_OFPT_FLOW_REMOVED:
3786 case OFPUTIL_OFPT_PORT_STATUS:
3787 case OFPUTIL_OFPT_BARRIER_REPLY:
3788 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3789 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3790 case OFPUTIL_OFPST_DESC_REPLY:
3791 case OFPUTIL_OFPST_FLOW_REPLY:
3792 case OFPUTIL_OFPST_QUEUE_REPLY:
3793 case OFPUTIL_OFPST_PORT_REPLY:
3794 case OFPUTIL_OFPST_TABLE_REPLY:
3795 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3796 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3797 case OFPUTIL_NXT_ROLE_REPLY:
3798 case OFPUTIL_NXT_FLOW_REMOVED:
3799 case OFPUTIL_NXT_PACKET_IN:
3800 case OFPUTIL_NXT_FLOW_MONITOR_PAUSED:
3801 case OFPUTIL_NXT_FLOW_MONITOR_RESUMED:
3802 case OFPUTIL_NXST_FLOW_REPLY:
3803 case OFPUTIL_NXST_AGGREGATE_REPLY:
3804 case OFPUTIL_NXST_FLOW_MONITOR_REPLY:
3806 return (oh->type == OFPT10_STATS_REQUEST ||
3807 oh->type == OFPT10_STATS_REPLY
3808 ? OFPERR_OFPBRC_BAD_STAT
3809 : OFPERR_OFPBRC_BAD_TYPE);
3814 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3816 int error = handle_openflow__(ofconn, ofp_msg);
3817 if (error && error != OFPROTO_POSTPONE) {
3818 ofconn_send_error(ofconn, ofp_msg->data, error);
3820 COVERAGE_INC(ofproto_recv_openflow);
3821 return error != OFPROTO_POSTPONE;
3824 /* Asynchronous operations. */
3826 /* Creates and returns a new ofopgroup that is not associated with any
3827 * OpenFlow connection.
3829 * The caller should add operations to the returned group with
3830 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3831 static struct ofopgroup *
3832 ofopgroup_create_unattached(struct ofproto *ofproto)
3834 struct ofopgroup *group = xzalloc(sizeof *group);
3835 group->ofproto = ofproto;
3836 list_init(&group->ofproto_node);
3837 list_init(&group->ops);
3838 list_init(&group->ofconn_node);
3842 /* Creates and returns a new ofopgroup for 'ofproto'.
3844 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3845 * connection. The 'request' and 'buffer_id' arguments are ignored.
3847 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3848 * If the ofopgroup eventually fails, then the error reply will include
3849 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3850 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3852 * The caller should add operations to the returned group with
3853 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3854 static struct ofopgroup *
3855 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3856 const struct ofp_header *request, uint32_t buffer_id)
3858 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3860 size_t request_len = ntohs(request->length);
3862 assert(ofconn_get_ofproto(ofconn) == ofproto);
3864 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3865 group->ofconn = ofconn;
3866 group->request = xmemdup(request, MIN(request_len, 64));
3867 group->buffer_id = buffer_id;
3872 /* Submits 'group' for processing.
3874 * If 'group' contains no operations (e.g. none were ever added, or all of the
3875 * ones that were added completed synchronously), then it is destroyed
3876 * immediately. Otherwise it is added to the ofproto's list of pending
3879 ofopgroup_submit(struct ofopgroup *group)
3881 if (!group->n_running) {
3882 ofopgroup_complete(group);
3884 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3885 group->ofproto->n_pending++;
3890 ofopgroup_complete(struct ofopgroup *group)
3892 struct ofproto *ofproto = group->ofproto;
3894 struct ofconn *abbrev_ofconn;
3895 ovs_be32 abbrev_xid;
3897 struct ofoperation *op, *next_op;
3900 assert(!group->n_running);
3903 LIST_FOR_EACH (op, group_node, &group->ops) {
3910 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
3911 LIST_FOR_EACH (op, group_node, &group->ops) {
3912 if (op->type != OFOPERATION_DELETE) {
3913 struct ofpbuf *packet;
3916 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3920 error = rule_execute(op->rule, in_port, packet);
3927 if (!error && !list_is_empty(&group->ofconn_node)) {
3928 abbrev_ofconn = group->ofconn;
3929 abbrev_xid = group->request->xid;
3931 abbrev_ofconn = NULL;
3932 abbrev_xid = htonl(0);
3934 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
3935 struct rule *rule = op->rule;
3937 if (!op->error && !ofproto_rule_is_hidden(rule)) {
3938 /* Check that we can just cast from ofoperation_type to
3939 * nx_flow_update_event. */
3940 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
3942 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
3944 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
3947 ofmonitor_report(ofproto->connmgr, rule,
3948 (enum nx_flow_update_event) op->type,
3949 op->reason, abbrev_ofconn, abbrev_xid);
3952 rule->pending = NULL;
3955 case OFOPERATION_ADD:
3957 ofproto_rule_destroy__(op->victim);
3958 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3959 == htons(VLAN_VID_MASK)) {
3960 if (ofproto->vlan_bitmap) {
3961 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3963 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3964 bitmap_set1(ofproto->vlan_bitmap, vid);
3965 ofproto->vlans_changed = true;
3968 ofproto->vlans_changed = true;
3972 oftable_substitute_rule(rule, op->victim);
3973 ofproto_rule_destroy__(rule);
3977 case OFOPERATION_DELETE:
3979 ofproto_rule_destroy__(rule);
3983 case OFOPERATION_MODIFY:
3985 rule->modified = time_msec();
3987 rule->flow_cookie = op->flow_cookie;
3989 free(rule->ofpacts);
3990 rule->ofpacts = op->ofpacts;
3991 rule->ofpacts_len = op->ofpacts_len;
3993 op->ofpacts_len = 0;
4002 ofoperation_destroy(op);
4005 ofmonitor_flush(ofproto->connmgr);
4007 if (!list_is_empty(&group->ofproto_node)) {
4008 assert(ofproto->n_pending > 0);
4009 ofproto->n_pending--;
4010 list_remove(&group->ofproto_node);
4012 if (!list_is_empty(&group->ofconn_node)) {
4013 list_remove(&group->ofconn_node);
4015 ofconn_send_error(group->ofconn, group->request, error);
4017 connmgr_retry(ofproto->connmgr);
4019 free(group->request);
4023 /* Initiates a new operation on 'rule', of the specified 'type', within
4024 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4026 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4027 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4029 * Returns the newly created ofoperation (which is also available as
4030 * rule->pending). */
4031 static struct ofoperation *
4032 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4033 enum ofoperation_type type,
4034 enum ofp_flow_removed_reason reason)
4036 struct ofproto *ofproto = group->ofproto;
4037 struct ofoperation *op;
4039 assert(!rule->pending);
4041 op = rule->pending = xzalloc(sizeof *op);
4043 list_push_back(&group->ops, &op->group_node);
4046 op->reason = reason;
4047 op->flow_cookie = rule->flow_cookie;
4051 if (type == OFOPERATION_DELETE) {
4052 hmap_insert(&ofproto->deletions, &op->hmap_node,
4053 cls_rule_hash(&rule->cr, rule->table_id));
4060 ofoperation_destroy(struct ofoperation *op)
4062 struct ofopgroup *group = op->group;
4065 op->rule->pending = NULL;
4067 if (op->type == OFOPERATION_DELETE) {
4068 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4070 list_remove(&op->group_node);
4075 /* Indicates that 'op' completed with status 'error', which is either 0 to
4076 * indicate success or an OpenFlow error code on failure.
4078 * If 'error' is 0, indicating success, the operation will be committed
4079 * permanently to the flow table. There is one interesting subcase:
4081 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4082 * the flow table (the "victim" rule) by a new one, then the caller must
4083 * have uninitialized any derived state in the victim rule, as in step 5 in
4084 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4085 * performs steps 6 and 7 for the victim rule, most notably by calling its
4086 * ->rule_dealloc() function.
4088 * If 'error' is nonzero, then generally the operation will be rolled back:
4090 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4091 * restores the original rule. The caller must have uninitialized any
4092 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4093 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4094 * and 7 for the new rule, calling its ->rule_dealloc() function.
4096 * - If 'op' is a "modify flow" operation, ofproto restores the original
4099 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4100 * allowed to fail. It must always succeed.
4102 * Please see the large comment in ofproto/ofproto-provider.h titled
4103 * "Asynchronous Operation Support" for more information. */
4105 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4107 struct ofopgroup *group = op->group;
4109 assert(op->rule->pending == op);
4110 assert(group->n_running > 0);
4111 assert(!error || op->type != OFOPERATION_DELETE);
4114 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4115 ofopgroup_complete(group);
4120 ofoperation_get_victim(struct ofoperation *op)
4122 assert(op->type == OFOPERATION_ADD);
4127 pick_datapath_id(const struct ofproto *ofproto)
4129 const struct ofport *port;
4131 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4133 uint8_t ea[ETH_ADDR_LEN];
4136 error = netdev_get_etheraddr(port->netdev, ea);
4138 return eth_addr_to_uint64(ea);
4140 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4141 ofproto->name, netdev_get_name(port->netdev),
4144 return ofproto->fallback_dpid;
4148 pick_fallback_dpid(void)
4150 uint8_t ea[ETH_ADDR_LEN];
4151 eth_addr_nicira_random(ea);
4152 return eth_addr_to_uint64(ea);
4155 /* Table overflow policy. */
4157 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4158 * is not configured to evict rules or if the table contains no evictable
4159 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4161 static struct rule *
4162 choose_rule_to_evict(struct oftable *table)
4164 struct eviction_group *evg;
4166 if (!table->eviction_fields) {
4170 /* In the common case, the outer and inner loops here will each be entered
4173 * - The inner loop normally "return"s in its first iteration. If the
4174 * eviction group has any evictable rules, then it always returns in
4177 * - The outer loop only iterates more than once if the largest eviction
4178 * group has no evictable rules.
4180 * - The outer loop can exit only if table's 'max_flows' is all filled up
4181 * by unevictable rules'. */
4182 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4185 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4186 if (rule->evictable) {
4195 /* Searches 'ofproto' for tables that have more flows than their configured
4196 * maximum and that have flow eviction enabled, and evicts as many flows as
4197 * necessary and currently feasible from them.
4199 * This triggers only when an OpenFlow table has N flows in it and then the
4200 * client configures a maximum number of flows less than N. */
4202 ofproto_evict(struct ofproto *ofproto)
4204 struct ofopgroup *group;
4205 struct oftable *table;
4207 group = ofopgroup_create_unattached(ofproto);
4208 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4209 while (classifier_count(&table->cls) > table->max_flows
4210 && table->eviction_fields) {
4213 rule = choose_rule_to_evict(table);
4214 if (!rule || rule->pending) {
4218 ofoperation_create(group, rule,
4219 OFOPERATION_DELETE, OFPRR_EVICTION);
4220 oftable_remove_rule(rule);
4221 ofproto->ofproto_class->rule_destruct(rule);
4224 ofopgroup_submit(group);
4227 /* Eviction groups. */
4229 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4230 * rules. The priority contains low-order random bits to ensure that eviction
4231 * groups with the same number of rules are prioritized randomly. */
4233 eviction_group_priority(size_t n_rules)
4235 uint16_t size = MIN(UINT16_MAX, n_rules);
4236 return (size << 16) | random_uint16();
4239 /* Updates 'evg', an eviction_group within 'table', following a change that
4240 * adds or removes rules in 'evg'. */
4242 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4244 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4245 eviction_group_priority(heap_count(&evg->rules)));
4248 /* Destroys 'evg', an eviction_group within 'table':
4250 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4251 * rules themselves, just removes them from the eviction group.)
4253 * - Removes 'evg' from 'table'.
4257 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4259 while (!heap_is_empty(&evg->rules)) {
4262 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4263 rule->eviction_group = NULL;
4265 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4266 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4267 heap_destroy(&evg->rules);
4271 /* Removes 'rule' from its eviction group, if any. */
4273 eviction_group_remove_rule(struct rule *rule)
4275 if (rule->eviction_group) {
4276 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4277 struct eviction_group *evg = rule->eviction_group;
4279 rule->eviction_group = NULL;
4280 heap_remove(&evg->rules, &rule->evg_node);
4281 if (heap_is_empty(&evg->rules)) {
4282 eviction_group_destroy(table, evg);
4284 eviction_group_resized(table, evg);
4289 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4290 * returns the hash value. */
4292 eviction_group_hash_rule(struct rule *rule)
4294 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4295 const struct mf_subfield *sf;
4298 hash = table->eviction_group_id_basis;
4299 for (sf = table->eviction_fields;
4300 sf < &table->eviction_fields[table->n_eviction_fields];
4303 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
4304 union mf_value value;
4306 mf_get_value(sf->field, &rule->cr.flow, &value);
4308 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4310 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4311 unsigned int start = sf->ofs + sf->n_bits;
4312 bitwise_zero(&value, sf->field->n_bytes, start,
4313 sf->field->n_bytes * 8 - start);
4315 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4317 hash = hash_int(hash, 0);
4324 /* Returns an eviction group within 'table' with the given 'id', creating one
4326 static struct eviction_group *
4327 eviction_group_find(struct oftable *table, uint32_t id)
4329 struct eviction_group *evg;
4331 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4335 evg = xmalloc(sizeof *evg);
4336 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4337 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4338 eviction_group_priority(0));
4339 heap_init(&evg->rules);
4344 /* Returns an eviction priority for 'rule'. The return value should be
4345 * interpreted so that higher priorities make a rule more attractive candidates
4348 rule_eviction_priority(struct rule *rule)
4350 long long int hard_expiration;
4351 long long int idle_expiration;
4352 long long int expiration;
4353 uint32_t expiration_offset;
4355 /* Calculate time of expiration. */
4356 hard_expiration = (rule->hard_timeout
4357 ? rule->modified + rule->hard_timeout * 1000
4359 idle_expiration = (rule->idle_timeout
4360 ? rule->used + rule->idle_timeout * 1000
4362 expiration = MIN(hard_expiration, idle_expiration);
4363 if (expiration == LLONG_MAX) {
4367 /* Calculate the time of expiration as a number of (approximate) seconds
4368 * after program startup.
4370 * This should work OK for program runs that last UINT32_MAX seconds or
4371 * less. Therefore, please restart OVS at least once every 136 years. */
4372 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4374 /* Invert the expiration offset because we're using a max-heap. */
4375 return UINT32_MAX - expiration_offset;
4378 /* Adds 'rule' to an appropriate eviction group for its oftable's
4379 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4380 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4383 * The caller must ensure that 'rule' is not already in an eviction group. */
4385 eviction_group_add_rule(struct rule *rule)
4387 struct ofproto *ofproto = rule->ofproto;
4388 struct oftable *table = &ofproto->tables[rule->table_id];
4390 if (table->eviction_fields
4391 && (rule->hard_timeout || rule->idle_timeout)) {
4392 struct eviction_group *evg;
4394 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4396 rule->eviction_group = evg;
4397 heap_insert(&evg->rules, &rule->evg_node,
4398 rule_eviction_priority(rule));
4399 eviction_group_resized(table, evg);
4405 /* Initializes 'table'. */
4407 oftable_init(struct oftable *table)
4409 memset(table, 0, sizeof *table);
4410 classifier_init(&table->cls);
4411 table->max_flows = UINT_MAX;
4414 /* Destroys 'table', including its classifier and eviction groups.
4416 * The caller is responsible for freeing 'table' itself. */
4418 oftable_destroy(struct oftable *table)
4420 assert(classifier_is_empty(&table->cls));
4421 oftable_disable_eviction(table);
4422 classifier_destroy(&table->cls);
4426 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4427 * string, then 'table' will use its default name.
4429 * This only affects the name exposed for a table exposed through the OpenFlow
4430 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4432 oftable_set_name(struct oftable *table, const char *name)
4434 if (name && name[0]) {
4435 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4436 if (!table->name || strncmp(name, table->name, len)) {
4438 table->name = xmemdup0(name, len);
4446 /* oftables support a choice of two policies when adding a rule would cause the
4447 * number of flows in the table to exceed the configured maximum number: either
4448 * they can refuse to add the new flow or they can evict some existing flow.
4449 * This function configures the former policy on 'table'. */
4451 oftable_disable_eviction(struct oftable *table)
4453 if (table->eviction_fields) {
4454 struct eviction_group *evg, *next;
4456 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4457 &table->eviction_groups_by_id) {
4458 eviction_group_destroy(table, evg);
4460 hmap_destroy(&table->eviction_groups_by_id);
4461 heap_destroy(&table->eviction_groups_by_size);
4463 free(table->eviction_fields);
4464 table->eviction_fields = NULL;
4465 table->n_eviction_fields = 0;
4469 /* oftables support a choice of two policies when adding a rule would cause the
4470 * number of flows in the table to exceed the configured maximum number: either
4471 * they can refuse to add the new flow or they can evict some existing flow.
4472 * This function configures the latter policy on 'table', with fairness based
4473 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4474 * 'n_fields' as 0 disables fairness.) */
4476 oftable_enable_eviction(struct oftable *table,
4477 const struct mf_subfield *fields, size_t n_fields)
4479 struct cls_cursor cursor;
4482 if (table->eviction_fields
4483 && n_fields == table->n_eviction_fields
4485 || !memcmp(fields, table->eviction_fields,
4486 n_fields * sizeof *fields))) {
4491 oftable_disable_eviction(table);
4493 table->n_eviction_fields = n_fields;
4494 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4496 table->eviction_group_id_basis = random_uint32();
4497 hmap_init(&table->eviction_groups_by_id);
4498 heap_init(&table->eviction_groups_by_size);
4500 cls_cursor_init(&cursor, &table->cls, NULL);
4501 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4502 eviction_group_add_rule(rule);
4506 /* Removes 'rule' from the oftable that contains it. */
4508 oftable_remove_rule(struct rule *rule)
4510 struct ofproto *ofproto = rule->ofproto;
4511 struct oftable *table = &ofproto->tables[rule->table_id];
4513 classifier_remove(&table->cls, &rule->cr);
4514 eviction_group_remove_rule(rule);
4517 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4518 * oftable that has an identical cls_rule. Returns the rule that was removed,
4519 * if any, and otherwise NULL. */
4520 static struct rule *
4521 oftable_replace_rule(struct rule *rule)
4523 struct ofproto *ofproto = rule->ofproto;
4524 struct oftable *table = &ofproto->tables[rule->table_id];
4525 struct rule *victim;
4527 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4529 eviction_group_remove_rule(victim);
4531 eviction_group_add_rule(rule);
4535 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4537 oftable_substitute_rule(struct rule *old, struct rule *new)
4540 oftable_replace_rule(new);
4542 oftable_remove_rule(old);
4546 /* unixctl commands. */
4549 ofproto_lookup(const char *name)
4551 struct ofproto *ofproto;
4553 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4555 if (!strcmp(ofproto->name, name)) {
4563 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4564 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4566 struct ofproto *ofproto;
4570 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4571 ds_put_format(&results, "%s\n", ofproto->name);
4573 unixctl_command_reply(conn, ds_cstr(&results));
4574 ds_destroy(&results);
4578 ofproto_unixctl_init(void)
4580 static bool registered;
4586 unixctl_command_register("ofproto/list", "", 0, 0,
4587 ofproto_unixctl_list, NULL);
4590 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4592 * This is deprecated. It is only for compatibility with broken device drivers
4593 * in old versions of Linux that do not properly support VLANs when VLAN
4594 * devices are not used. When broken device drivers are no longer in
4595 * widespread use, we will delete these interfaces. */
4597 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4598 * (exactly) by an OpenFlow rule in 'ofproto'. */
4600 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4602 const struct oftable *oftable;
4604 free(ofproto->vlan_bitmap);
4605 ofproto->vlan_bitmap = bitmap_allocate(4096);
4606 ofproto->vlans_changed = false;
4608 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4609 const struct cls_table *table;
4611 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4612 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4613 == htons(VLAN_VID_MASK)) {
4614 const struct cls_rule *rule;
4616 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4617 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4618 bitmap_set1(vlan_bitmap, vid);
4619 bitmap_set1(ofproto->vlan_bitmap, vid);
4626 /* Returns true if new VLANs have come into use by the flow table since the
4627 * last call to ofproto_get_vlan_usage().
4629 * We don't track when old VLANs stop being used. */
4631 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4633 return ofproto->vlans_changed;
4636 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4637 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4638 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4639 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4640 * then the VLAN device is un-enslaved. */
4642 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4643 uint16_t realdev_ofp_port, int vid)
4645 struct ofport *ofport;
4648 assert(vlandev_ofp_port != realdev_ofp_port);
4650 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4652 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4653 ofproto->name, vlandev_ofp_port);
4657 if (!ofproto->ofproto_class->set_realdev) {
4658 if (!vlandev_ofp_port) {
4661 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4665 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4667 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4668 ofproto->name, vlandev_ofp_port,
4669 netdev_get_name(ofport->netdev), strerror(error));