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. */
91 /* Data needed to send OpenFlow reply on failure or to send a buffered
94 * If list_is_empty(ofconn_node) then this ofopgroup never had an
95 * associated ofconn or its ofconn's connection dropped after it initiated
96 * the operation. In the latter case 'ofconn' is a wild pointer that
97 * refers to freed memory, so the 'ofconn' member must be used only if
98 * !list_is_empty(ofconn_node).
100 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
101 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
102 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
103 uint32_t buffer_id; /* Buffer id from original request. */
104 int error; /* 0 if no error yet, otherwise error code. */
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_destroy(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. */
121 struct rule *victim; /* OFOPERATION_ADD: Replaced rule. */
122 struct ofpact *ofpacts; /* OFOPERATION_MODIFY: Replaced actions. */
123 size_t ofpacts_len; /* OFOPERATION_MODIFY: Bytes of ofpacts. */
124 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
127 static struct ofoperation *ofoperation_create(struct ofopgroup *,
129 enum ofoperation_type);
130 static void ofoperation_destroy(struct ofoperation *);
133 static void oftable_init(struct oftable *);
134 static void oftable_destroy(struct oftable *);
136 static void oftable_set_name(struct oftable *, const char *name);
138 static void oftable_disable_eviction(struct oftable *);
139 static void oftable_enable_eviction(struct oftable *,
140 const struct mf_subfield *fields,
143 static void oftable_remove_rule(struct rule *);
144 static struct rule *oftable_replace_rule(struct rule *);
145 static void oftable_substitute_rule(struct rule *old, struct rule *new);
147 /* A set of rules within a single OpenFlow table (oftable) that have the same
148 * values for the oftable's eviction_fields. A rule to be evicted, when one is
149 * needed, is taken from the eviction group that contains the greatest number
152 * An oftable owns any number of eviction groups, each of which contains any
155 * Membership in an eviction group is imprecise, based on the hash of the
156 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
157 * That is, if two rules have different eviction_fields, but those
158 * eviction_fields hash to the same value, then they will belong to the same
159 * eviction_group anyway.
161 * (When eviction is not enabled on an oftable, we don't track any eviction
162 * groups, to save time and space.) */
163 struct eviction_group {
164 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
165 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
166 struct heap rules; /* Contains "struct rule"s. */
169 static struct rule *choose_rule_to_evict(struct oftable *);
170 static void ofproto_evict(struct ofproto *);
171 static uint32_t rule_eviction_priority(struct rule *);
174 static void ofport_destroy__(struct ofport *);
175 static void ofport_destroy(struct ofport *);
177 static void update_port(struct ofproto *, const char *devname);
178 static int init_ports(struct ofproto *);
179 static void reinit_ports(struct ofproto *);
182 static void ofproto_rule_destroy__(struct rule *);
183 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
184 static bool rule_is_modifiable(const struct rule *);
185 static bool rule_is_hidden(const struct rule *);
188 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
189 const struct ofputil_flow_mod *,
190 const struct ofp_header *);
191 static void delete_flow__(struct rule *, struct ofopgroup *);
192 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
193 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
194 const struct ofputil_flow_mod *,
195 const struct ofp_header *);
198 static uint64_t pick_datapath_id(const struct ofproto *);
199 static uint64_t pick_fallback_dpid(void);
200 static void ofproto_destroy__(struct ofproto *);
201 static void update_mtu(struct ofproto *, struct ofport *);
204 static void ofproto_unixctl_init(void);
206 /* All registered ofproto classes, in probe order. */
207 static const struct ofproto_class **ofproto_classes;
208 static size_t n_ofproto_classes;
209 static size_t allocated_ofproto_classes;
211 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
212 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
214 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
217 ofproto_initialize(void)
223 ofproto_class_register(&ofproto_dpif_class);
227 /* 'type' should be a normalized datapath type, as returned by
228 * ofproto_normalize_type(). Returns the corresponding ofproto_class
229 * structure, or a null pointer if there is none registered for 'type'. */
230 static const struct ofproto_class *
231 ofproto_class_find__(const char *type)
235 ofproto_initialize();
236 for (i = 0; i < n_ofproto_classes; i++) {
237 const struct ofproto_class *class = ofproto_classes[i];
242 class->enumerate_types(&types);
243 found = sset_contains(&types, type);
244 sset_destroy(&types);
250 VLOG_WARN("unknown datapath type %s", type);
254 /* Registers a new ofproto class. After successful registration, new ofprotos
255 * of that type can be created using ofproto_create(). */
257 ofproto_class_register(const struct ofproto_class *new_class)
261 for (i = 0; i < n_ofproto_classes; i++) {
262 if (ofproto_classes[i] == new_class) {
267 if (n_ofproto_classes >= allocated_ofproto_classes) {
268 ofproto_classes = x2nrealloc(ofproto_classes,
269 &allocated_ofproto_classes,
270 sizeof *ofproto_classes);
272 ofproto_classes[n_ofproto_classes++] = new_class;
276 /* Unregisters a datapath provider. 'type' must have been previously
277 * registered and not currently be in use by any ofprotos. After
278 * unregistration new datapaths of that type cannot be opened using
279 * ofproto_create(). */
281 ofproto_class_unregister(const struct ofproto_class *class)
285 for (i = 0; i < n_ofproto_classes; i++) {
286 if (ofproto_classes[i] == class) {
287 for (i++; i < n_ofproto_classes; i++) {
288 ofproto_classes[i - 1] = ofproto_classes[i];
294 VLOG_WARN("attempted to unregister an ofproto class that is not "
299 /* Clears 'types' and enumerates all registered ofproto types into it. The
300 * caller must first initialize the sset. */
302 ofproto_enumerate_types(struct sset *types)
306 ofproto_initialize();
307 for (i = 0; i < n_ofproto_classes; i++) {
308 ofproto_classes[i]->enumerate_types(types);
312 /* Returns the fully spelled out name for the given ofproto 'type'.
314 * Normalized type string can be compared with strcmp(). Unnormalized type
315 * string might be the same even if they have different spellings. */
317 ofproto_normalize_type(const char *type)
319 return type && type[0] ? type : "system";
322 /* Clears 'names' and enumerates the names of all known created ofprotos with
323 * the given 'type'. The caller must first initialize the sset. Returns 0 if
324 * successful, otherwise a positive errno value.
326 * Some kinds of datapaths might not be practically enumerable. This is not
327 * considered an error. */
329 ofproto_enumerate_names(const char *type, struct sset *names)
331 const struct ofproto_class *class = ofproto_class_find__(type);
332 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
336 ofproto_create(const char *datapath_name, const char *datapath_type,
337 struct ofproto **ofprotop)
339 const struct ofproto_class *class;
340 struct ofproto *ofproto;
345 ofproto_initialize();
346 ofproto_unixctl_init();
348 datapath_type = ofproto_normalize_type(datapath_type);
349 class = ofproto_class_find__(datapath_type);
351 VLOG_WARN("could not create datapath %s of unknown type %s",
352 datapath_name, datapath_type);
356 ofproto = class->alloc();
358 VLOG_ERR("failed to allocate datapath %s of type %s",
359 datapath_name, datapath_type);
364 memset(ofproto, 0, sizeof *ofproto);
365 ofproto->ofproto_class = class;
366 ofproto->name = xstrdup(datapath_name);
367 ofproto->type = xstrdup(datapath_type);
368 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
369 hash_string(ofproto->name, 0));
370 ofproto->datapath_id = 0;
371 ofproto_set_flow_eviction_threshold(ofproto,
372 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
373 ofproto->forward_bpdu = false;
374 ofproto->fallback_dpid = pick_fallback_dpid();
375 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
376 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
377 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
378 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
379 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
380 ofproto->frag_handling = OFPC_FRAG_NORMAL;
381 hmap_init(&ofproto->ports);
382 shash_init(&ofproto->port_by_name);
383 ofproto->tables = NULL;
384 ofproto->n_tables = 0;
385 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
386 ofproto->state = S_OPENFLOW;
387 list_init(&ofproto->pending);
388 ofproto->n_pending = 0;
389 hmap_init(&ofproto->deletions);
390 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
391 ofproto->first_op = ofproto->last_op = LLONG_MIN;
392 ofproto->next_op_report = LLONG_MAX;
393 ofproto->op_backoff = LLONG_MIN;
394 ofproto->vlan_bitmap = NULL;
395 ofproto->vlans_changed = false;
396 ofproto->min_mtu = INT_MAX;
398 error = ofproto->ofproto_class->construct(ofproto);
400 VLOG_ERR("failed to open datapath %s: %s",
401 datapath_name, strerror(error));
402 ofproto_destroy__(ofproto);
406 assert(ofproto->n_tables);
408 ofproto->datapath_id = pick_datapath_id(ofproto);
416 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
418 struct oftable *table;
420 assert(!ofproto->n_tables);
421 assert(n_tables >= 1 && n_tables <= 255);
423 ofproto->n_tables = n_tables;
424 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
425 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
431 ofproto_get_datapath_id(const struct ofproto *ofproto)
433 return ofproto->datapath_id;
437 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
439 uint64_t old_dpid = p->datapath_id;
440 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
441 if (p->datapath_id != old_dpid) {
442 /* Force all active connections to reconnect, since there is no way to
443 * notify a controller that the datapath ID has changed. */
444 ofproto_reconnect_controllers(p);
449 ofproto_set_controllers(struct ofproto *p,
450 const struct ofproto_controller *controllers,
451 size_t n_controllers)
453 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
457 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
459 connmgr_set_fail_mode(p->connmgr, fail_mode);
462 /* Drops the connections between 'ofproto' and all of its controllers, forcing
463 * them to reconnect. */
465 ofproto_reconnect_controllers(struct ofproto *ofproto)
467 connmgr_reconnect(ofproto->connmgr);
470 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
471 * in-band control should guarantee access, in the same way that in-band
472 * control guarantees access to OpenFlow controllers. */
474 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
475 const struct sockaddr_in *extras, size_t n)
477 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
480 /* Sets the OpenFlow queue used by flows set up by in-band control on
481 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
482 * flows will use the default queue. */
484 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
486 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
489 /* Sets the number of flows at which eviction from the kernel flow table
492 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
494 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
495 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
497 ofproto->flow_eviction_threshold = threshold;
501 /* If forward_bpdu is true, the NORMAL action will forward frames with
502 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
503 * the NORMAL action will drop these frames. */
505 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
507 bool old_val = ofproto->forward_bpdu;
508 ofproto->forward_bpdu = forward_bpdu;
509 if (old_val != ofproto->forward_bpdu) {
510 if (ofproto->ofproto_class->forward_bpdu_changed) {
511 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
516 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
517 * 'idle_time', in seconds. */
519 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
521 if (ofproto->ofproto_class->set_mac_idle_time) {
522 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
527 ofproto_set_desc(struct ofproto *p,
528 const char *mfr_desc, const char *hw_desc,
529 const char *sw_desc, const char *serial_desc,
532 struct ofp_desc_stats *ods;
535 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
536 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
537 p->name, sizeof ods->mfr_desc);
540 p->mfr_desc = xstrdup(mfr_desc);
543 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
544 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
545 p->name, sizeof ods->hw_desc);
548 p->hw_desc = xstrdup(hw_desc);
551 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
552 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
553 p->name, sizeof ods->sw_desc);
556 p->sw_desc = xstrdup(sw_desc);
559 if (strlen(serial_desc) >= sizeof ods->serial_num) {
560 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
561 "bytes", p->name, sizeof ods->serial_num);
563 free(p->serial_desc);
564 p->serial_desc = xstrdup(serial_desc);
567 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
568 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
569 p->name, sizeof ods->dp_desc);
572 p->dp_desc = xstrdup(dp_desc);
577 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
579 return connmgr_set_snoops(ofproto->connmgr, snoops);
583 ofproto_set_netflow(struct ofproto *ofproto,
584 const struct netflow_options *nf_options)
586 if (nf_options && sset_is_empty(&nf_options->collectors)) {
590 if (ofproto->ofproto_class->set_netflow) {
591 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
593 return nf_options ? EOPNOTSUPP : 0;
598 ofproto_set_sflow(struct ofproto *ofproto,
599 const struct ofproto_sflow_options *oso)
601 if (oso && sset_is_empty(&oso->targets)) {
605 if (ofproto->ofproto_class->set_sflow) {
606 return ofproto->ofproto_class->set_sflow(ofproto, oso);
608 return oso ? EOPNOTSUPP : 0;
612 /* Spanning Tree Protocol (STP) configuration. */
614 /* Configures STP on 'ofproto' using the settings defined in 's'. If
615 * 's' is NULL, disables STP.
617 * Returns 0 if successful, otherwise a positive errno value. */
619 ofproto_set_stp(struct ofproto *ofproto,
620 const struct ofproto_stp_settings *s)
622 return (ofproto->ofproto_class->set_stp
623 ? ofproto->ofproto_class->set_stp(ofproto, s)
627 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
628 * 'enabled' member of 's' is false, then the other members are not
631 * Returns 0 if successful, otherwise a positive errno value. */
633 ofproto_get_stp_status(struct ofproto *ofproto,
634 struct ofproto_stp_status *s)
636 return (ofproto->ofproto_class->get_stp_status
637 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
641 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
642 * in 's'. The caller is responsible for assigning STP port numbers
643 * (using the 'port_num' member in the range of 1 through 255, inclusive)
644 * and ensuring there are no duplicates. If the 's' is NULL, then STP
645 * is disabled on the port.
647 * Returns 0 if successful, otherwise a positive errno value.*/
649 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
650 const struct ofproto_port_stp_settings *s)
652 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
654 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
655 ofproto->name, ofp_port);
659 return (ofproto->ofproto_class->set_stp_port
660 ? ofproto->ofproto_class->set_stp_port(ofport, s)
664 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
665 * 's'. If the 'enabled' member in 's' is false, then the other members
666 * are not meaningful.
668 * Returns 0 if successful, otherwise a positive errno value.*/
670 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
671 struct ofproto_port_stp_status *s)
673 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
675 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
676 "port %"PRIu16, ofproto->name, ofp_port);
680 return (ofproto->ofproto_class->get_stp_port_status
681 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
685 /* Queue DSCP configuration. */
687 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
688 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
689 * to implement QoS. Instead, it is used to implement features which require
690 * knowledge of what queues exist on a port, and some basic information about
693 * Returns 0 if successful, otherwise a positive errno value. */
695 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
696 const struct ofproto_port_queue *queues,
699 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
702 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
703 ofproto->name, ofp_port);
707 return (ofproto->ofproto_class->set_queues
708 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
712 /* Connectivity Fault Management configuration. */
714 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
716 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
718 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
719 if (ofport && ofproto->ofproto_class->set_cfm) {
720 ofproto->ofproto_class->set_cfm(ofport, NULL);
724 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
725 * basic configuration from the configuration members in 'cfm', and the remote
726 * maintenance point ID from remote_mpid. Ignores the statistics members of
729 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
731 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
732 const struct cfm_settings *s)
734 struct ofport *ofport;
737 ofport = ofproto_get_port(ofproto, ofp_port);
739 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
740 ofproto->name, ofp_port);
744 /* XXX: For configuration simplicity, we only support one remote_mpid
745 * outside of the CFM module. It's not clear if this is the correct long
746 * term solution or not. */
747 error = (ofproto->ofproto_class->set_cfm
748 ? ofproto->ofproto_class->set_cfm(ofport, s)
751 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
752 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
757 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
758 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
759 * 0 if LACP partner information is not current (generally indicating a
760 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
762 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
764 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
765 return (ofport && ofproto->ofproto_class->port_is_lacp_current
766 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
772 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
773 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
774 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
775 * configuration plus, if there is more than one slave, a bonding
778 * If 'aux' is already registered then this function updates its configuration
779 * to 's'. Otherwise, this function registers a new bundle.
781 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
784 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
785 const struct ofproto_bundle_settings *s)
787 return (ofproto->ofproto_class->bundle_set
788 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
792 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
793 * If no such bundle has been registered, this has no effect. */
795 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
797 return ofproto_bundle_register(ofproto, aux, NULL);
801 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
802 * If 'aux' is already registered then this function updates its configuration
803 * to 's'. Otherwise, this function registers a new mirror. */
805 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
806 const struct ofproto_mirror_settings *s)
808 return (ofproto->ofproto_class->mirror_set
809 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
813 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
814 * If no mirror has been registered, this has no effect. */
816 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
818 return ofproto_mirror_register(ofproto, aux, NULL);
821 /* Retrieves statistics from mirror associated with client data pointer
822 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
823 * 'bytes', respectively. If a particular counters is not supported,
824 * the appropriate argument is set to UINT64_MAX. */
826 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
827 uint64_t *packets, uint64_t *bytes)
829 if (!ofproto->ofproto_class->mirror_get_stats) {
830 *packets = *bytes = UINT64_MAX;
834 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
838 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
839 * which all packets are flooded, instead of using MAC learning. If
840 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
842 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
845 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
847 return (ofproto->ofproto_class->set_flood_vlans
848 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
852 /* Returns true if 'aux' is a registered bundle that is currently in use as the
853 * output for a mirror. */
855 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
857 return (ofproto->ofproto_class->is_mirror_output_bundle
858 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
862 /* Configuration of OpenFlow tables. */
864 /* Returns the number of OpenFlow tables in 'ofproto'. */
866 ofproto_get_n_tables(const struct ofproto *ofproto)
868 return ofproto->n_tables;
871 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
872 * settings from 's'. 'table_id' must be in the range 0 through the number of
873 * OpenFlow tables in 'ofproto' minus 1, inclusive.
875 * For read-only tables, only the name may be configured. */
877 ofproto_configure_table(struct ofproto *ofproto, int table_id,
878 const struct ofproto_table_settings *s)
880 struct oftable *table;
882 assert(table_id >= 0 && table_id < ofproto->n_tables);
883 table = &ofproto->tables[table_id];
885 oftable_set_name(table, s->name);
887 if (table->flags & OFTABLE_READONLY) {
892 oftable_enable_eviction(table, s->groups, s->n_groups);
894 oftable_disable_eviction(table);
897 table->max_flows = s->max_flows;
898 if (classifier_count(&table->cls) > table->max_flows
899 && table->eviction_fields) {
900 /* 'table' contains more flows than allowed. We might not be able to
901 * evict them right away because of the asynchronous nature of flow
902 * table changes. Schedule eviction for later. */
903 switch (ofproto->state) {
905 ofproto->state = S_EVICT;
909 /* We're already deleting flows, nothing more to do. */
916 ofproto_has_snoops(const struct ofproto *ofproto)
918 return connmgr_has_snoops(ofproto->connmgr);
922 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
924 connmgr_get_snoops(ofproto->connmgr, snoops);
928 ofproto_flush__(struct ofproto *ofproto)
930 struct ofopgroup *group;
931 struct oftable *table;
933 if (ofproto->ofproto_class->flush) {
934 ofproto->ofproto_class->flush(ofproto);
937 group = ofopgroup_create_unattached(ofproto);
938 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
939 struct rule *rule, *next_rule;
940 struct cls_cursor cursor;
942 if (table->flags & OFTABLE_HIDDEN) {
946 cls_cursor_init(&cursor, &table->cls, NULL);
947 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
948 if (!rule->pending) {
949 ofoperation_create(group, rule, OFOPERATION_DELETE);
950 oftable_remove_rule(rule);
951 ofproto->ofproto_class->rule_destruct(rule);
955 ofopgroup_submit(group);
959 ofproto_destroy__(struct ofproto *ofproto)
961 struct oftable *table;
963 assert(list_is_empty(&ofproto->pending));
964 assert(!ofproto->n_pending);
966 connmgr_destroy(ofproto->connmgr);
968 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
971 free(ofproto->mfr_desc);
972 free(ofproto->hw_desc);
973 free(ofproto->sw_desc);
974 free(ofproto->serial_desc);
975 free(ofproto->dp_desc);
976 hmap_destroy(&ofproto->ports);
977 shash_destroy(&ofproto->port_by_name);
979 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
980 oftable_destroy(table);
982 free(ofproto->tables);
984 hmap_destroy(&ofproto->deletions);
986 free(ofproto->vlan_bitmap);
988 ofproto->ofproto_class->dealloc(ofproto);
992 ofproto_destroy(struct ofproto *p)
994 struct ofport *ofport, *next_ofport;
1001 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1002 ofport_destroy(ofport);
1005 p->ofproto_class->destruct(p);
1006 ofproto_destroy__(p);
1009 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1010 * kernel datapath, for example, this destroys the datapath in the kernel, and
1011 * with the netdev-based datapath, it tears down the data structures that
1012 * represent the datapath.
1014 * The datapath should not be currently open as an ofproto. */
1016 ofproto_delete(const char *name, const char *type)
1018 const struct ofproto_class *class = ofproto_class_find__(type);
1019 return (!class ? EAFNOSUPPORT
1020 : !class->del ? EACCES
1021 : class->del(type, name));
1025 process_port_change(struct ofproto *ofproto, int error, char *devname)
1027 if (error == ENOBUFS) {
1028 reinit_ports(ofproto);
1029 } else if (!error) {
1030 update_port(ofproto, devname);
1036 ofproto_run(struct ofproto *p)
1038 struct sset changed_netdevs;
1039 const char *changed_netdev;
1040 struct ofport *ofport;
1043 error = p->ofproto_class->run(p);
1044 if (error && error != EAGAIN) {
1045 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1048 if (p->ofproto_class->port_poll) {
1051 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1052 process_port_change(p, error, devname);
1056 /* Update OpenFlow port status for any port whose netdev has changed.
1058 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1059 * destroyed, so it's not safe to update ports directly from the
1060 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1061 * need this two-phase approach. */
1062 sset_init(&changed_netdevs);
1063 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1064 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1065 if (ofport->change_seq != change_seq) {
1066 ofport->change_seq = change_seq;
1067 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1070 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1071 update_port(p, changed_netdev);
1073 sset_destroy(&changed_netdevs);
1077 connmgr_run(p->connmgr, handle_openflow);
1081 connmgr_run(p->connmgr, NULL);
1083 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1084 p->state = S_OPENFLOW;
1089 connmgr_run(p->connmgr, NULL);
1091 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1092 connmgr_flushed(p->connmgr);
1093 p->state = S_OPENFLOW;
1101 if (time_msec() >= p->next_op_report) {
1102 long long int ago = (time_msec() - p->first_op) / 1000;
1103 long long int interval = (p->last_op - p->first_op) / 1000;
1107 ds_put_format(&s, "%d flow_mods ",
1108 p->n_add + p->n_delete + p->n_modify);
1109 if (interval == ago) {
1110 ds_put_format(&s, "in the last %lld s", ago);
1111 } else if (interval) {
1112 ds_put_format(&s, "in the %lld s starting %lld s ago",
1115 ds_put_format(&s, "%lld s ago", ago);
1118 ds_put_cstr(&s, " (");
1120 ds_put_format(&s, "%d adds, ", p->n_add);
1123 ds_put_format(&s, "%d deletes, ", p->n_delete);
1126 ds_put_format(&s, "%d modifications, ", p->n_modify);
1129 ds_put_char(&s, ')');
1131 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1134 p->n_add = p->n_delete = p->n_modify = 0;
1135 p->next_op_report = LLONG_MAX;
1141 /* Performs periodic activity required by 'ofproto' that needs to be done
1142 * with the least possible latency.
1144 * It makes sense to call this function a couple of times per poll loop, to
1145 * provide a significant performance boost on some benchmarks with the
1146 * ofproto-dpif implementation. */
1148 ofproto_run_fast(struct ofproto *p)
1152 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1153 if (error && error != EAGAIN) {
1154 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1155 p->name, strerror(error));
1161 ofproto_wait(struct ofproto *p)
1163 struct ofport *ofport;
1165 p->ofproto_class->wait(p);
1166 if (p->ofproto_class->port_poll_wait) {
1167 p->ofproto_class->port_poll_wait(p);
1170 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1171 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1172 poll_immediate_wake();
1178 connmgr_wait(p->connmgr, true);
1183 connmgr_wait(p->connmgr, false);
1184 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1185 poll_immediate_wake();
1192 ofproto_is_alive(const struct ofproto *p)
1194 return connmgr_has_controllers(p->connmgr);
1197 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1198 * memory_report(). */
1200 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1202 const struct oftable *table;
1203 unsigned int n_rules;
1205 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1206 simap_increase(usage, "ops",
1207 ofproto->n_pending + hmap_count(&ofproto->deletions));
1210 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1211 n_rules += classifier_count(&table->cls);
1213 simap_increase(usage, "rules", n_rules);
1215 if (ofproto->ofproto_class->get_memory_usage) {
1216 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1219 connmgr_get_memory_usage(ofproto->connmgr, usage);
1223 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1226 connmgr_get_controller_info(ofproto->connmgr, info);
1230 ofproto_free_ofproto_controller_info(struct shash *info)
1232 connmgr_free_controller_info(info);
1235 /* Makes a deep copy of 'old' into 'port'. */
1237 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1239 port->name = xstrdup(old->name);
1240 port->type = xstrdup(old->type);
1241 port->ofp_port = old->ofp_port;
1244 /* Frees memory allocated to members of 'ofproto_port'.
1246 * Do not call this function on an ofproto_port obtained from
1247 * ofproto_port_dump_next(): that function retains ownership of the data in the
1250 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1252 free(ofproto_port->name);
1253 free(ofproto_port->type);
1256 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1258 * This function provides no status indication. An error status for the entire
1259 * dump operation is provided when it is completed by calling
1260 * ofproto_port_dump_done().
1263 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1264 const struct ofproto *ofproto)
1266 dump->ofproto = ofproto;
1267 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1271 /* Attempts to retrieve another port from 'dump', which must have been created
1272 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1273 * 'port' and returns true. On failure, returns false.
1275 * Failure might indicate an actual error or merely that the last port has been
1276 * dumped. An error status for the entire dump operation is provided when it
1277 * is completed by calling ofproto_port_dump_done().
1279 * The ofproto owns the data stored in 'port'. It will remain valid until at
1280 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1281 * ofproto_port_dump_done(). */
1283 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1284 struct ofproto_port *port)
1286 const struct ofproto *ofproto = dump->ofproto;
1292 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1295 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1301 /* Completes port table dump operation 'dump', which must have been created
1302 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1303 * error-free, otherwise a positive errno value describing the problem. */
1305 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1307 const struct ofproto *ofproto = dump->ofproto;
1309 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1312 return dump->error == EOF ? 0 : dump->error;
1315 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1316 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1317 * is non-null). On failure, returns a positive errno value and sets
1318 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1320 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1321 uint16_t *ofp_portp)
1326 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1328 update_port(ofproto, netdev_get_name(netdev));
1331 *ofp_portp = error ? OFPP_NONE : ofp_port;
1336 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1337 * initializes '*port' appropriately; on failure, returns a positive errno
1340 * The caller owns the data in 'ofproto_port' and must free it with
1341 * ofproto_port_destroy() when it is no longer needed. */
1343 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1344 struct ofproto_port *port)
1348 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1350 memset(port, 0, sizeof *port);
1355 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1356 * Returns 0 if successful, otherwise a positive errno. */
1358 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1360 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1361 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1364 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1365 if (!error && ofport) {
1366 /* 'name' is the netdev's name and update_port() is going to close the
1367 * netdev. Just in case update_port() refers to 'name' after it
1368 * destroys 'ofport', make a copy of it around the update_port()
1370 char *devname = xstrdup(name);
1371 update_port(ofproto, devname);
1377 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1378 * performs the 'n_actions' actions in 'actions'. The new flow will not
1381 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1382 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1383 * controllers; otherwise, it will be hidden.
1385 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1387 * This is a helper function for in-band control and fail-open. */
1389 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1390 const struct ofpact *ofpacts, size_t ofpacts_len)
1392 const struct rule *rule;
1394 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1395 &ofproto->tables[0].cls, cls_rule));
1396 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1397 ofpacts, ofpacts_len)) {
1398 struct ofputil_flow_mod fm;
1400 memset(&fm, 0, sizeof fm);
1402 fm.buffer_id = UINT32_MAX;
1403 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1404 fm.ofpacts_len = ofpacts_len;
1405 add_flow(ofproto, NULL, &fm, NULL);
1410 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1411 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1412 * operation cannot be initiated now but may be retried later.
1414 * This is a helper function for in-band control and fail-open. */
1416 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1418 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1421 /* Searches for a rule with matching criteria exactly equal to 'target' in
1422 * ofproto's table 0 and, if it finds one, deletes it.
1424 * This is a helper function for in-band control and fail-open. */
1426 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1430 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1431 &ofproto->tables[0].cls, target));
1433 /* No such rule -> success. */
1435 } else if (rule->pending) {
1436 /* An operation on the rule is already pending -> failure.
1437 * Caller must retry later if it's important. */
1440 /* Initiate deletion -> success. */
1441 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1442 ofoperation_create(group, rule, OFOPERATION_DELETE);
1443 oftable_remove_rule(rule);
1444 ofproto->ofproto_class->rule_destruct(rule);
1445 ofopgroup_submit(group);
1451 /* Starts the process of deleting all of the flows from all of ofproto's flow
1452 * tables and then reintroducing the flows required by in-band control and
1453 * fail-open. The process will complete in a later call to ofproto_run(). */
1455 ofproto_flush_flows(struct ofproto *ofproto)
1457 COVERAGE_INC(ofproto_flush);
1458 ofproto->state = S_FLUSH;
1462 reinit_ports(struct ofproto *p)
1464 struct ofproto_port_dump dump;
1465 struct sset devnames;
1466 struct ofport *ofport;
1467 struct ofproto_port ofproto_port;
1468 const char *devname;
1470 COVERAGE_INC(ofproto_reinit_ports);
1472 sset_init(&devnames);
1473 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1474 sset_add(&devnames, netdev_get_name(ofport->netdev));
1476 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1477 sset_add(&devnames, ofproto_port.name);
1480 SSET_FOR_EACH (devname, &devnames) {
1481 update_port(p, devname);
1483 sset_destroy(&devnames);
1486 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1487 * pointer if the netdev cannot be opened. On success, also fills in
1489 static struct netdev *
1490 ofport_open(const struct ofproto *ofproto,
1491 const struct ofproto_port *ofproto_port,
1492 struct ofputil_phy_port *pp)
1494 enum netdev_flags flags;
1495 struct netdev *netdev;
1498 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1500 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1501 "cannot be opened (%s)",
1503 ofproto_port->name, ofproto_port->ofp_port,
1504 ofproto_port->name, strerror(error));
1508 pp->port_no = ofproto_port->ofp_port;
1509 netdev_get_etheraddr(netdev, pp->hw_addr);
1510 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1511 netdev_get_flags(netdev, &flags);
1512 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1513 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1514 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1515 &pp->supported, &pp->peer);
1516 pp->curr_speed = netdev_features_to_bps(pp->curr);
1517 pp->max_speed = netdev_features_to_bps(pp->supported);
1522 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1523 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1526 ofport_equal(const struct ofputil_phy_port *a,
1527 const struct ofputil_phy_port *b)
1529 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1530 && a->state == b->state
1531 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1532 && a->curr == b->curr
1533 && a->advertised == b->advertised
1534 && a->supported == b->supported
1535 && a->peer == b->peer
1536 && a->curr_speed == b->curr_speed
1537 && a->max_speed == b->max_speed);
1540 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1541 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1542 * one with the same name or port number). */
1544 ofport_install(struct ofproto *p,
1545 struct netdev *netdev, const struct ofputil_phy_port *pp)
1547 const char *netdev_name = netdev_get_name(netdev);
1548 struct ofport *ofport;
1551 /* Create ofport. */
1552 ofport = p->ofproto_class->port_alloc();
1557 ofport->ofproto = p;
1558 ofport->netdev = netdev;
1559 ofport->change_seq = netdev_change_seq(netdev);
1561 ofport->ofp_port = pp->port_no;
1563 /* Add port to 'p'. */
1564 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1565 shash_add(&p->port_by_name, netdev_name, ofport);
1567 update_mtu(p, ofport);
1569 /* Let the ofproto_class initialize its private data. */
1570 error = p->ofproto_class->port_construct(ofport);
1574 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1578 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1579 p->name, netdev_name, strerror(error));
1581 ofport_destroy__(ofport);
1583 netdev_close(netdev);
1587 /* Removes 'ofport' from 'p' and destroys it. */
1589 ofport_remove(struct ofport *ofport)
1591 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1593 ofport_destroy(ofport);
1596 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1599 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1601 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1603 ofport_remove(port);
1607 /* Updates 'port' with new 'pp' description.
1609 * Does not handle a name or port number change. The caller must implement
1610 * such a change as a delete followed by an add. */
1612 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1614 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1615 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1616 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1617 port->pp.state = pp->state;
1618 port->pp.curr = pp->curr;
1619 port->pp.advertised = pp->advertised;
1620 port->pp.supported = pp->supported;
1621 port->pp.peer = pp->peer;
1622 port->pp.curr_speed = pp->curr_speed;
1623 port->pp.max_speed = pp->max_speed;
1625 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1628 /* Update OpenFlow 'state' in 'port' and notify controller. */
1630 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1632 if (port->pp.state != state) {
1633 port->pp.state = state;
1634 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1640 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1642 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1644 if (port->ofproto->ofproto_class->set_realdev) {
1645 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1647 if (port->ofproto->ofproto_class->set_stp_port) {
1648 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1650 if (port->ofproto->ofproto_class->set_cfm) {
1651 port->ofproto->ofproto_class->set_cfm(port, NULL);
1653 if (port->ofproto->ofproto_class->bundle_remove) {
1654 port->ofproto->ofproto_class->bundle_remove(port);
1660 ofport_destroy__(struct ofport *port)
1662 struct ofproto *ofproto = port->ofproto;
1663 const char *name = netdev_get_name(port->netdev);
1665 hmap_remove(&ofproto->ports, &port->hmap_node);
1666 shash_delete(&ofproto->port_by_name,
1667 shash_find(&ofproto->port_by_name, name));
1669 netdev_close(port->netdev);
1670 ofproto->ofproto_class->port_dealloc(port);
1674 ofport_destroy(struct ofport *port)
1677 port->ofproto->ofproto_class->port_destruct(port);
1678 ofport_destroy__(port);
1683 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1685 struct ofport *port;
1687 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1688 hash_int(ofp_port, 0), &ofproto->ports) {
1689 if (port->ofp_port == ofp_port) {
1697 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1699 struct ofproto *ofproto = port->ofproto;
1702 if (ofproto->ofproto_class->port_get_stats) {
1703 error = ofproto->ofproto_class->port_get_stats(port, stats);
1712 update_port(struct ofproto *ofproto, const char *name)
1714 struct ofproto_port ofproto_port;
1715 struct ofputil_phy_port pp;
1716 struct netdev *netdev;
1717 struct ofport *port;
1719 COVERAGE_INC(ofproto_update_port);
1721 /* Fetch 'name''s location and properties from the datapath. */
1722 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1723 ? ofport_open(ofproto, &ofproto_port, &pp)
1726 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1727 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1728 struct netdev *old_netdev = port->netdev;
1730 /* 'name' hasn't changed location. Any properties changed? */
1731 if (!ofport_equal(&port->pp, &pp)) {
1732 ofport_modified(port, &pp);
1735 update_mtu(ofproto, port);
1737 /* Install the newly opened netdev in case it has changed.
1738 * Don't close the old netdev yet in case port_modified has to
1739 * remove a retained reference to it.*/
1740 port->netdev = netdev;
1741 port->change_seq = netdev_change_seq(netdev);
1743 if (port->ofproto->ofproto_class->port_modified) {
1744 port->ofproto->ofproto_class->port_modified(port);
1747 netdev_close(old_netdev);
1749 /* If 'port' is nonnull then its name differs from 'name' and thus
1750 * we should delete it. If we think there's a port named 'name'
1751 * then its port number must be wrong now so delete it too. */
1753 ofport_remove(port);
1755 ofport_remove_with_name(ofproto, name);
1756 ofport_install(ofproto, netdev, &pp);
1759 /* Any port named 'name' is gone now. */
1760 ofport_remove_with_name(ofproto, name);
1762 ofproto_port_destroy(&ofproto_port);
1766 init_ports(struct ofproto *p)
1768 struct ofproto_port_dump dump;
1769 struct ofproto_port ofproto_port;
1771 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1772 uint16_t ofp_port = ofproto_port.ofp_port;
1773 if (ofproto_get_port(p, ofp_port)) {
1774 VLOG_WARN_RL(&rl, "%s: ignoring duplicate port %"PRIu16" "
1775 "in datapath", p->name, ofp_port);
1776 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1777 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1778 p->name, ofproto_port.name);
1780 struct ofputil_phy_port pp;
1781 struct netdev *netdev;
1783 netdev = ofport_open(p, &ofproto_port, &pp);
1785 ofport_install(p, netdev, &pp);
1793 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1794 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1796 find_min_mtu(struct ofproto *p)
1798 struct ofport *ofport;
1801 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1802 struct netdev *netdev = ofport->netdev;
1805 /* Skip any internal ports, since that's what we're trying to
1807 if (!strcmp(netdev_get_type(netdev), "internal")) {
1811 if (netdev_get_mtu(netdev, &dev_mtu)) {
1814 if (!mtu || dev_mtu < mtu) {
1819 return mtu ? mtu: ETH_PAYLOAD_MAX;
1822 /* Update MTU of all datapath devices on 'p' to the minimum of the
1823 * non-datapath ports in event of 'port' added or changed. */
1825 update_mtu(struct ofproto *p, struct ofport *port)
1827 struct ofport *ofport;
1828 struct netdev *netdev = port->netdev;
1829 int dev_mtu, old_min;
1831 if (netdev_get_mtu(netdev, &dev_mtu)) {
1835 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1836 if (dev_mtu > p->min_mtu) {
1837 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1838 dev_mtu = p->min_mtu;
1841 port->mtu = dev_mtu;
1845 /* For non-internal port find new min mtu. */
1846 old_min = p->min_mtu;
1847 port->mtu = dev_mtu;
1848 p->min_mtu = find_min_mtu(p);
1849 if (p->min_mtu == old_min) {
1853 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1854 struct netdev *netdev = ofport->netdev;
1856 if (!strcmp(netdev_get_type(netdev), "internal")) {
1857 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1858 ofport->mtu = p->min_mtu;
1865 ofproto_rule_destroy__(struct rule *rule)
1868 free(rule->ofpacts);
1869 rule->ofproto->ofproto_class->rule_dealloc(rule);
1873 /* This function allows an ofproto implementation to destroy any rules that
1874 * remain when its ->destruct() function is called. The caller must have
1875 * already uninitialized any derived members of 'rule' (step 5 described in the
1876 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1877 * This function implements steps 6 and 7.
1879 * This function should only be called from an ofproto implementation's
1880 * ->destruct() function. It is not suitable elsewhere. */
1882 ofproto_rule_destroy(struct rule *rule)
1884 assert(!rule->pending);
1885 oftable_remove_rule(rule);
1886 ofproto_rule_destroy__(rule);
1889 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1890 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1892 rule_has_out_port(const struct rule *rule, uint16_t port)
1894 return (port == OFPP_NONE
1895 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
1898 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1899 * statistics appropriately. 'packet' must have at least sizeof(struct
1900 * ofp_packet_in) bytes of headroom.
1902 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1903 * with statistics for 'packet' either way.
1905 * Takes ownership of 'packet'. */
1907 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1911 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1913 flow_extract(packet, 0, 0, in_port, &flow);
1914 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1917 /* Returns true if 'rule' should be hidden from the controller.
1919 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1920 * (e.g. by in-band control) and are intentionally hidden from the
1923 rule_is_hidden(const struct rule *rule)
1925 return rule->cr.priority > UINT16_MAX;
1928 static enum oftable_flags
1929 rule_get_flags(const struct rule *rule)
1931 return rule->ofproto->tables[rule->table_id].flags;
1935 rule_is_modifiable(const struct rule *rule)
1937 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1941 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1943 ofconn_send_reply(ofconn, make_echo_reply(oh));
1948 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1950 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1951 struct ofputil_switch_features features;
1952 struct ofport *port;
1956 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1958 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1960 features.datapath_id = ofproto->datapath_id;
1961 features.n_buffers = pktbuf_capacity();
1962 features.n_tables = ofproto->n_tables;
1963 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1964 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1966 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
1969 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
1971 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1972 ofputil_put_switch_features_port(&port->pp, b);
1975 ofconn_send_reply(ofconn, b);
1980 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1982 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1983 struct ofp_switch_config *osc;
1984 enum ofp_config_flags flags;
1988 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1989 flags = ofproto->frag_handling;
1990 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1991 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1993 osc->flags = htons(flags);
1994 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1995 ofconn_send_reply(ofconn, buf);
2001 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2003 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2004 uint16_t flags = ntohs(osc->flags);
2006 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2007 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2008 enum ofp_config_flags cur = ofproto->frag_handling;
2009 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2011 assert((cur & OFPC_FRAG_MASK) == cur);
2013 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2014 ofproto->frag_handling = next;
2016 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2018 ofputil_frag_handling_to_string(next));
2022 ofconn_set_invalid_ttl_to_controller(ofconn,
2023 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2025 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2030 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2031 * error message code for the caller to propagate upward. Otherwise, returns
2034 * The log message mentions 'msg_type'. */
2036 reject_slave_controller(struct ofconn *ofconn)
2038 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2039 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2040 return OFPERR_OFPBRC_EPERM;
2047 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
2049 struct ofproto *p = ofconn_get_ofproto(ofconn);
2050 struct ofputil_packet_out po;
2051 struct ofpbuf *payload;
2052 uint64_t ofpacts_stub[1024 / 8];
2053 struct ofpbuf ofpacts;
2057 COVERAGE_INC(ofproto_packet_out);
2059 error = reject_slave_controller(ofconn);
2064 /* Decode message. */
2065 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2066 error = ofputil_decode_packet_out(&po, opo, &ofpacts);
2068 goto exit_free_ofpacts;
2072 if (po.buffer_id != UINT32_MAX) {
2073 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2074 if (error || !payload) {
2075 goto exit_free_ofpacts;
2078 payload = xmalloc(sizeof *payload);
2079 ofpbuf_use_const(payload, po.packet, po.packet_len);
2082 /* Send out packet. */
2083 flow_extract(payload, 0, 0, po.in_port, &flow);
2084 error = p->ofproto_class->packet_out(p, payload, &flow,
2085 po.ofpacts, po.ofpacts_len);
2086 ofpbuf_delete(payload);
2089 ofpbuf_uninit(&ofpacts);
2095 update_port_config(struct ofport *port,
2096 enum ofputil_port_config config,
2097 enum ofputil_port_config mask)
2099 enum ofputil_port_config old_config = port->pp.config;
2100 enum ofputil_port_config toggle;
2102 toggle = (config ^ port->pp.config) & mask;
2103 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2104 if (config & OFPUTIL_PC_PORT_DOWN) {
2105 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2107 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2109 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2112 port->pp.config ^= toggle;
2113 if (port->pp.config != old_config) {
2114 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2119 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2121 struct ofproto *p = ofconn_get_ofproto(ofconn);
2122 struct ofputil_port_mod pm;
2123 struct ofport *port;
2126 error = reject_slave_controller(ofconn);
2131 error = ofputil_decode_port_mod(oh, &pm);
2136 port = ofproto_get_port(p, pm.port_no);
2138 return OFPERR_OFPPMFC_BAD_PORT;
2139 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2140 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2142 update_port_config(port, pm.config, pm.mask);
2144 netdev_set_advertisements(port->netdev, pm.advertise);
2151 handle_desc_stats_request(struct ofconn *ofconn,
2152 const struct ofp_stats_msg *request)
2154 struct ofproto *p = ofconn_get_ofproto(ofconn);
2155 struct ofp_desc_stats *ods;
2158 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2159 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2160 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2161 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2162 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2163 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2164 ofconn_send_reply(ofconn, msg);
2170 handle_table_stats_request(struct ofconn *ofconn,
2171 const struct ofp_stats_msg *request)
2173 struct ofproto *p = ofconn_get_ofproto(ofconn);
2174 struct ofp_table_stats *ots;
2178 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2180 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2181 for (i = 0; i < p->n_tables; i++) {
2182 ots[i].table_id = i;
2183 sprintf(ots[i].name, "table%zu", i);
2184 ots[i].wildcards = htonl(OFPFW10_ALL);
2185 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2186 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2189 p->ofproto_class->get_tables(p, ots);
2191 for (i = 0; i < p->n_tables; i++) {
2192 const struct oftable *table = &p->tables[i];
2195 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2198 if (table->max_flows < ntohl(ots[i].max_entries)) {
2199 ots[i].max_entries = htonl(table->max_flows);
2203 ofconn_send_reply(ofconn, msg);
2208 append_port_stat(struct ofport *port, struct list *replies)
2210 struct netdev_stats stats;
2211 struct ofp_port_stats *ops;
2213 /* Intentionally ignore return value, since errors will set
2214 * 'stats' to all-1s, which is correct for OpenFlow, and
2215 * netdev_get_stats() will log errors. */
2216 ofproto_port_get_stats(port, &stats);
2218 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2219 ops->port_no = htons(port->pp.port_no);
2220 memset(ops->pad, 0, sizeof ops->pad);
2221 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2222 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2223 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2224 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2225 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2226 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2227 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2228 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2229 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2230 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2231 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2232 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2236 handle_port_stats_request(struct ofconn *ofconn,
2237 const struct ofp_port_stats_request *psr)
2239 struct ofproto *p = ofconn_get_ofproto(ofconn);
2240 struct ofport *port;
2241 struct list replies;
2243 ofputil_start_stats_reply(&psr->osm, &replies);
2244 if (psr->port_no != htons(OFPP_NONE)) {
2245 port = ofproto_get_port(p, ntohs(psr->port_no));
2247 append_port_stat(port, &replies);
2250 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2251 append_port_stat(port, &replies);
2255 ofconn_send_replies(ofconn, &replies);
2260 handle_port_desc_stats_request(struct ofconn *ofconn,
2261 const struct ofp_stats_msg *osm)
2263 struct ofproto *p = ofconn_get_ofproto(ofconn);
2264 struct ofport *port;
2265 struct list replies;
2267 ofputil_start_stats_reply(osm, &replies);
2269 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2270 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2271 &port->pp, &replies);
2274 ofconn_send_replies(ofconn, &replies);
2279 calc_flow_duration__(long long int start, long long int now,
2280 uint32_t *sec, uint32_t *nsec)
2282 long long int msecs = now - start;
2283 *sec = msecs / 1000;
2284 *nsec = (msecs % 1000) * (1000 * 1000);
2287 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2288 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2290 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2292 return (table_id == 0xff || table_id < ofproto->n_tables
2294 : OFPERR_NXBRC_BAD_TABLE_ID);
2298 static struct oftable *
2299 next_visible_table(struct ofproto *ofproto, uint8_t table_id)
2301 struct oftable *table;
2303 for (table = &ofproto->tables[table_id];
2304 table < &ofproto->tables[ofproto->n_tables];
2306 if (!(table->flags & OFTABLE_HIDDEN)) {
2314 static struct oftable *
2315 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2317 if (table_id == 0xff) {
2318 return next_visible_table(ofproto, 0);
2319 } else if (table_id < ofproto->n_tables) {
2320 return &ofproto->tables[table_id];
2326 static struct oftable *
2327 next_matching_table(struct ofproto *ofproto,
2328 struct oftable *table, uint8_t table_id)
2330 return (table_id == 0xff
2331 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2335 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2337 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2338 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2340 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2341 * only once, for that table. (This can be used to access tables marked
2344 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2345 * entered at all. (Perhaps you should have validated TABLE_ID with
2346 * check_table_id().)
2348 * All parameters are evaluated multiple times.
2350 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2351 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2353 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2355 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2356 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2357 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2360 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2361 * to 'out_port' are included.
2363 * Hidden rules are always omitted.
2365 * Returns 0 on success, otherwise an OpenFlow error code. */
2367 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2368 const struct cls_rule *match,
2369 ovs_be64 cookie, ovs_be64 cookie_mask,
2370 uint16_t out_port, struct list *rules)
2372 struct oftable *table;
2375 error = check_table_id(ofproto, table_id);
2381 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2382 struct cls_cursor cursor;
2385 cls_cursor_init(&cursor, &table->cls, match);
2386 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2387 if (rule->pending) {
2388 return OFPROTO_POSTPONE;
2390 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2391 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2392 list_push_back(rules, &rule->ofproto_node);
2399 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2400 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2401 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2404 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2405 * to 'out_port' are included.
2407 * Hidden rules are always omitted.
2409 * Returns 0 on success, otherwise an OpenFlow error code. */
2411 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2412 const struct cls_rule *match,
2413 ovs_be64 cookie, ovs_be64 cookie_mask,
2414 uint16_t out_port, struct list *rules)
2416 struct oftable *table;
2419 error = check_table_id(ofproto, table_id);
2425 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2428 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2431 if (rule->pending) {
2432 return OFPROTO_POSTPONE;
2434 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2435 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2436 list_push_back(rules, &rule->ofproto_node);
2443 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2444 * forced into the range of a uint16_t. */
2446 age_secs(long long int age_ms)
2448 return (age_ms < 0 ? 0
2449 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2450 : (unsigned int) age_ms / 1000);
2454 handle_flow_stats_request(struct ofconn *ofconn,
2455 const struct ofp_stats_msg *osm)
2457 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2458 struct ofputil_flow_stats_request fsr;
2459 struct list replies;
2464 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2469 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2470 fsr.cookie, fsr.cookie_mask,
2471 fsr.out_port, &rules);
2476 ofputil_start_stats_reply(osm, &replies);
2477 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2478 long long int now = time_msec();
2479 struct ofputil_flow_stats fs;
2482 fs.cookie = rule->flow_cookie;
2483 fs.table_id = rule->table_id;
2484 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2486 fs.idle_timeout = rule->idle_timeout;
2487 fs.hard_timeout = rule->hard_timeout;
2488 fs.idle_age = age_secs(now - rule->used);
2489 fs.hard_age = age_secs(now - rule->modified);
2490 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2492 fs.ofpacts = rule->ofpacts;
2493 fs.ofpacts_len = rule->ofpacts_len;
2494 ofputil_append_flow_stats_reply(&fs, &replies);
2496 ofconn_send_replies(ofconn, &replies);
2502 flow_stats_ds(struct rule *rule, struct ds *results)
2504 uint64_t packet_count, byte_count;
2506 rule->ofproto->ofproto_class->rule_get_stats(rule,
2507 &packet_count, &byte_count);
2509 if (rule->table_id != 0) {
2510 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2512 ds_put_format(results, "duration=%llds, ",
2513 (time_msec() - rule->created) / 1000);
2514 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2515 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2516 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2517 cls_rule_format(&rule->cr, results);
2518 ds_put_char(results, ',');
2519 if (rule->ofpacts_len > 0) {
2520 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2522 ds_put_cstr(results, "drop");
2524 ds_put_cstr(results, "\n");
2527 /* Adds a pretty-printed description of all flows to 'results', including
2528 * hidden flows (e.g., set up by in-band control). */
2530 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2532 struct oftable *table;
2534 OFPROTO_FOR_EACH_TABLE (table, p) {
2535 struct cls_cursor cursor;
2538 cls_cursor_init(&cursor, &table->cls, NULL);
2539 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2540 flow_stats_ds(rule, results);
2545 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2546 * '*engine_type' and '*engine_id', respectively. */
2548 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2549 uint8_t *engine_type, uint8_t *engine_id)
2551 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2554 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2555 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2556 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2557 * and -1 if CFM is not enabled on 'port'. */
2559 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2561 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2562 return (ofport && ofproto->ofproto_class->get_cfm_fault
2563 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2567 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2568 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2569 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2570 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2572 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2573 uint16_t ofp_port, const uint64_t **rmps,
2576 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2580 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2581 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2586 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2587 * integer value between 0 and 100 to indicate the health of the port as a
2588 * percentage which is the average of cfm health of all the remote_mpids or
2589 * returns -1 if CFM is not enabled on 'ofport'. */
2591 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2593 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2594 return (ofport && ofproto->ofproto_class->get_cfm_health
2595 ? ofproto->ofproto_class->get_cfm_health(ofport)
2600 handle_aggregate_stats_request(struct ofconn *ofconn,
2601 const struct ofp_stats_msg *osm)
2603 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2604 struct ofputil_flow_stats_request request;
2605 struct ofputil_aggregate_stats stats;
2606 bool unknown_packets, unknown_bytes;
2607 struct ofpbuf *reply;
2612 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2617 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2618 request.cookie, request.cookie_mask,
2619 request.out_port, &rules);
2624 memset(&stats, 0, sizeof stats);
2625 unknown_packets = unknown_bytes = false;
2626 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2627 uint64_t packet_count;
2628 uint64_t byte_count;
2630 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2633 if (packet_count == UINT64_MAX) {
2634 unknown_packets = true;
2636 stats.packet_count += packet_count;
2639 if (byte_count == UINT64_MAX) {
2640 unknown_bytes = true;
2642 stats.byte_count += byte_count;
2647 if (unknown_packets) {
2648 stats.packet_count = UINT64_MAX;
2650 if (unknown_bytes) {
2651 stats.byte_count = UINT64_MAX;
2654 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2655 ofconn_send_reply(ofconn, reply);
2660 struct queue_stats_cbdata {
2661 struct ofport *ofport;
2662 struct list replies;
2666 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2667 const struct netdev_queue_stats *stats)
2669 struct ofp_queue_stats *reply;
2671 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2672 reply->port_no = htons(cbdata->ofport->pp.port_no);
2673 memset(reply->pad, 0, sizeof reply->pad);
2674 reply->queue_id = htonl(queue_id);
2675 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2676 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2677 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2681 handle_queue_stats_dump_cb(uint32_t queue_id,
2682 struct netdev_queue_stats *stats,
2685 struct queue_stats_cbdata *cbdata = cbdata_;
2687 put_queue_stats(cbdata, queue_id, stats);
2691 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2692 struct queue_stats_cbdata *cbdata)
2694 cbdata->ofport = port;
2695 if (queue_id == OFPQ_ALL) {
2696 netdev_dump_queue_stats(port->netdev,
2697 handle_queue_stats_dump_cb, cbdata);
2699 struct netdev_queue_stats stats;
2701 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2702 put_queue_stats(cbdata, queue_id, &stats);
2704 return OFPERR_OFPQOFC_BAD_QUEUE;
2711 handle_queue_stats_request(struct ofconn *ofconn,
2712 const struct ofp_queue_stats_request *qsr)
2714 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2715 struct queue_stats_cbdata cbdata;
2716 unsigned int port_no;
2717 struct ofport *port;
2721 COVERAGE_INC(ofproto_queue_req);
2723 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2725 port_no = ntohs(qsr->port_no);
2726 queue_id = ntohl(qsr->queue_id);
2727 if (port_no == OFPP_ALL) {
2728 error = OFPERR_OFPQOFC_BAD_QUEUE;
2729 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2730 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2735 port = ofproto_get_port(ofproto, port_no);
2737 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2738 : OFPERR_OFPQOFC_BAD_PORT);
2741 ofconn_send_replies(ofconn, &cbdata.replies);
2743 ofpbuf_list_delete(&cbdata.replies);
2750 is_flow_deletion_pending(const struct ofproto *ofproto,
2751 const struct cls_rule *cls_rule,
2754 if (!hmap_is_empty(&ofproto->deletions)) {
2755 struct ofoperation *op;
2757 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2758 cls_rule_hash(cls_rule, table_id),
2759 &ofproto->deletions) {
2760 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2769 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2770 * in which no matching flow already exists in the flow table.
2772 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2773 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2774 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2775 * initiated now but may be retried later.
2777 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2778 * ownership remains with the caller.
2780 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2783 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2784 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2786 struct oftable *table;
2787 struct ofopgroup *group;
2788 struct rule *victim;
2792 error = check_table_id(ofproto, fm->table_id);
2798 if (fm->table_id == 0xff) {
2800 if (ofproto->ofproto_class->rule_choose_table) {
2801 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2806 assert(table_id < ofproto->n_tables);
2807 table = &ofproto->tables[table_id];
2809 table = &ofproto->tables[0];
2811 } else if (fm->table_id < ofproto->n_tables) {
2812 table = &ofproto->tables[fm->table_id];
2814 return OFPERR_NXFMFC_BAD_TABLE_ID;
2817 if (table->flags & OFTABLE_READONLY) {
2818 return OFPERR_OFPBRC_EPERM;
2821 /* Check for overlap, if requested. */
2822 if (fm->flags & OFPFF_CHECK_OVERLAP
2823 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2824 return OFPERR_OFPFMFC_OVERLAP;
2827 /* Serialize against pending deletion. */
2828 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2829 return OFPROTO_POSTPONE;
2832 /* Allocate new rule. */
2833 rule = ofproto->ofproto_class->rule_alloc();
2835 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2836 ofproto->name, strerror(error));
2839 rule->ofproto = ofproto;
2841 rule->pending = NULL;
2842 rule->flow_cookie = fm->new_cookie;
2843 rule->created = rule->modified = rule->used = time_msec();
2844 rule->idle_timeout = fm->idle_timeout;
2845 rule->hard_timeout = fm->hard_timeout;
2846 rule->table_id = table - ofproto->tables;
2847 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2848 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2849 rule->ofpacts_len = fm->ofpacts_len;
2850 rule->evictable = true;
2851 rule->eviction_group = NULL;
2853 /* Insert new rule. */
2854 victim = oftable_replace_rule(rule);
2855 if (victim && !rule_is_modifiable(victim)) {
2856 error = OFPERR_OFPBRC_EPERM;
2857 } else if (victim && victim->pending) {
2858 error = OFPROTO_POSTPONE;
2860 struct ofoperation *op;
2863 if (classifier_count(&table->cls) > table->max_flows) {
2866 was_evictable = rule->evictable;
2867 rule->evictable = false;
2868 evict = choose_rule_to_evict(table);
2869 rule->evictable = was_evictable;
2872 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2874 } else if (evict->pending) {
2875 error = OFPROTO_POSTPONE;
2882 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2883 op = ofoperation_create(group, rule, OFOPERATION_ADD);
2884 op->victim = victim;
2886 error = ofproto->ofproto_class->rule_construct(rule);
2888 ofoperation_destroy(rule->pending);
2890 delete_flow__(evict, group);
2892 ofopgroup_submit(group);
2896 /* Back out if an error occurred. */
2898 oftable_substitute_rule(rule, victim);
2899 ofproto_rule_destroy__(rule);
2904 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2906 /* Modifies the rules listed in 'rules', changing their actions to match those
2909 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2912 * Returns 0 on success, otherwise an OpenFlow error code. */
2914 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2915 const struct ofputil_flow_mod *fm,
2916 const struct ofp_header *request, struct list *rules)
2918 struct ofopgroup *group;
2922 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2923 error = OFPERR_OFPBRC_EPERM;
2924 LIST_FOR_EACH (rule, ofproto_node, rules) {
2925 if (rule_is_modifiable(rule)) {
2926 /* At least one rule is modifiable, don't report EPERM error. */
2932 if (!ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2933 rule->ofpacts, rule->ofpacts_len)) {
2934 struct ofoperation *op;
2936 op = ofoperation_create(group, rule, OFOPERATION_MODIFY);
2937 op->ofpacts = rule->ofpacts;
2938 op->ofpacts_len = rule->ofpacts_len;
2939 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2940 rule->ofpacts_len = fm->ofpacts_len;
2941 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2943 rule->modified = time_msec();
2945 if (fm->new_cookie != htonll(UINT64_MAX)) {
2946 rule->flow_cookie = fm->new_cookie;
2949 ofopgroup_submit(group);
2954 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2957 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2960 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2961 const struct ofputil_flow_mod *fm,
2962 const struct ofp_header *request)
2967 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2968 fm->cookie, fm->cookie_mask,
2972 } else if (list_is_empty(&rules)) {
2973 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2975 return modify_flows__(ofproto, ofconn, fm, request, &rules);
2979 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2982 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2985 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2986 const struct ofputil_flow_mod *fm,
2987 const struct ofp_header *request)
2992 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2993 fm->cookie, fm->cookie_mask,
2998 } else if (list_is_empty(&rules)) {
2999 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
3001 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3002 fm, request, &rules)
3007 /* OFPFC_DELETE implementation. */
3010 delete_flow__(struct rule *rule, struct ofopgroup *group)
3012 struct ofproto *ofproto = rule->ofproto;
3014 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3016 ofoperation_create(group, rule, OFOPERATION_DELETE);
3017 oftable_remove_rule(rule);
3018 ofproto->ofproto_class->rule_destruct(rule);
3021 /* Deletes the rules listed in 'rules'.
3023 * Returns 0 on success, otherwise an OpenFlow error code. */
3025 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3026 const struct ofp_header *request, struct list *rules)
3028 struct rule *rule, *next;
3029 struct ofopgroup *group;
3031 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3032 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3033 delete_flow__(rule, group);
3035 ofopgroup_submit(group);
3040 /* Implements OFPFC_DELETE. */
3042 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3043 const struct ofputil_flow_mod *fm,
3044 const struct ofp_header *request)
3049 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3050 fm->cookie, fm->cookie_mask,
3051 fm->out_port, &rules);
3052 return (error ? error
3053 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3058 /* Implements OFPFC_DELETE_STRICT. */
3060 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3061 const struct ofputil_flow_mod *fm,
3062 const struct ofp_header *request)
3067 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3068 fm->cookie, fm->cookie_mask,
3069 fm->out_port, &rules);
3070 return (error ? error
3071 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3077 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3079 struct ofputil_flow_removed fr;
3081 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
3086 fr.cookie = rule->flow_cookie;
3088 calc_flow_duration__(rule->created, time_msec(),
3089 &fr.duration_sec, &fr.duration_nsec);
3090 fr.idle_timeout = rule->idle_timeout;
3091 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3094 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3098 ofproto_rule_update_used(struct rule *rule, long long int used)
3100 if (used > rule->used) {
3101 struct eviction_group *evg = rule->eviction_group;
3105 heap_change(&evg->rules, &rule->evg_node,
3106 rule_eviction_priority(rule));
3111 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3112 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3115 * ofproto implementation ->run() functions should use this function to expire
3116 * OpenFlow flows. */
3118 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3120 struct ofproto *ofproto = rule->ofproto;
3121 struct ofopgroup *group;
3123 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3125 ofproto_rule_send_removed(rule, reason);
3127 group = ofopgroup_create_unattached(ofproto);
3128 ofoperation_create(group, rule, OFOPERATION_DELETE);
3129 oftable_remove_rule(rule);
3130 ofproto->ofproto_class->rule_destruct(rule);
3131 ofopgroup_submit(group);
3135 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3137 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3138 struct ofputil_flow_mod fm;
3139 uint64_t ofpacts_stub[1024 / 8];
3140 struct ofpbuf ofpacts;
3144 error = reject_slave_controller(ofconn);
3149 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3150 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3153 goto exit_free_ofpacts;
3156 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3157 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3158 if (fm.flags & OFPFF_EMERG) {
3159 /* We do not support the emergency flow cache. It will hopefully get
3160 * dropped from OpenFlow in the near future. There is no good error
3161 * code, so just state that the flow table is full. */
3162 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
3164 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3167 goto exit_free_ofpacts;
3170 /* Record the operation for logging a summary report. */
3171 switch (fm.command) {
3177 case OFPFC_MODIFY_STRICT:
3178 ofproto->n_modify++;
3182 case OFPFC_DELETE_STRICT:
3183 ofproto->n_delete++;
3188 if (ofproto->next_op_report == LLONG_MAX) {
3189 ofproto->first_op = now;
3190 ofproto->next_op_report = MAX(now + 10 * 1000,
3191 ofproto->op_backoff);
3192 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3194 ofproto->last_op = now;
3197 ofpbuf_uninit(&ofpacts);
3203 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3204 const struct ofputil_flow_mod *fm,
3205 const struct ofp_header *oh)
3207 if (ofproto->n_pending >= 50) {
3208 assert(!list_is_empty(&ofproto->pending));
3209 return OFPROTO_POSTPONE;
3212 switch (fm->command) {
3214 return add_flow(ofproto, ofconn, fm, oh);
3217 return modify_flows_loose(ofproto, ofconn, fm, oh);
3219 case OFPFC_MODIFY_STRICT:
3220 return modify_flow_strict(ofproto, ofconn, fm, oh);
3223 return delete_flows_loose(ofproto, ofconn, fm, oh);
3225 case OFPFC_DELETE_STRICT:
3226 return delete_flow_strict(ofproto, ofconn, fm, oh);
3229 if (fm->command > 0xff) {
3230 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3231 "flow_mod_table_id extension is not enabled",
3234 return OFPERR_OFPFMFC_BAD_COMMAND;
3239 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3241 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3242 struct nx_role_request *reply;
3246 role = ntohl(nrr->role);
3247 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3248 && role != NX_ROLE_SLAVE) {
3249 return OFPERR_OFPRRFC_BAD_ROLE;
3252 if (ofconn_get_role(ofconn) != role
3253 && ofconn_has_pending_opgroups(ofconn)) {
3254 return OFPROTO_POSTPONE;
3257 ofconn_set_role(ofconn, role);
3259 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3260 reply->role = htonl(role);
3261 ofconn_send_reply(ofconn, buf);
3267 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3268 const struct ofp_header *oh)
3270 const struct nx_flow_mod_table_id *msg
3271 = (const struct nx_flow_mod_table_id *) oh;
3272 enum ofputil_protocol cur, next;
3274 cur = ofconn_get_protocol(ofconn);
3275 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3276 ofconn_set_protocol(ofconn, next);
3282 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3284 const struct nx_set_flow_format *msg
3285 = (const struct nx_set_flow_format *) oh;
3286 enum ofputil_protocol cur, next;
3287 enum ofputil_protocol next_base;
3289 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3291 return OFPERR_OFPBRC_EPERM;
3294 cur = ofconn_get_protocol(ofconn);
3295 next = ofputil_protocol_set_base(cur, next_base);
3296 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3297 /* Avoid sending async messages in surprising protocol. */
3298 return OFPROTO_POSTPONE;
3301 ofconn_set_protocol(ofconn, next);
3306 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3307 const struct ofp_header *oh)
3309 const struct nx_set_packet_in_format *msg;
3312 msg = (const struct nx_set_packet_in_format *) oh;
3313 format = ntohl(msg->format);
3314 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3315 return OFPERR_OFPBRC_EPERM;
3318 if (format != ofconn_get_packet_in_format(ofconn)
3319 && ofconn_has_pending_opgroups(ofconn)) {
3320 /* Avoid sending async message in surprsing packet in format. */
3321 return OFPROTO_POSTPONE;
3324 ofconn_set_packet_in_format(ofconn, format);
3329 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3331 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3332 uint32_t master[OAM_N_TYPES];
3333 uint32_t slave[OAM_N_TYPES];
3335 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3336 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3337 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3339 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3340 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3341 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3343 ofconn_set_async_config(ofconn, master, slave);
3344 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3345 !ofconn_get_miss_send_len(ofconn)) {
3346 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3353 handle_nxt_set_controller_id(struct ofconn *ofconn,
3354 const struct ofp_header *oh)
3356 const struct nx_controller_id *nci;
3358 nci = (const struct nx_controller_id *) oh;
3359 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3360 return OFPERR_NXBRC_MUST_BE_ZERO;
3363 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3368 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3372 if (ofconn_has_pending_opgroups(ofconn)) {
3373 return OFPROTO_POSTPONE;
3376 make_openflow_xid(sizeof *oh, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3377 ofconn_send_reply(ofconn, buf);
3382 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3384 const struct ofp_header *oh = msg->data;
3385 const struct ofputil_msg_type *type;
3388 error = ofputil_decode_msg_type(oh, &type);
3393 switch (ofputil_msg_type_code(type)) {
3394 /* OpenFlow requests. */
3395 case OFPUTIL_OFPT_ECHO_REQUEST:
3396 return handle_echo_request(ofconn, oh);
3398 case OFPUTIL_OFPT_FEATURES_REQUEST:
3399 return handle_features_request(ofconn, oh);
3401 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3402 return handle_get_config_request(ofconn, oh);
3404 case OFPUTIL_OFPT_SET_CONFIG:
3405 return handle_set_config(ofconn, msg->data);
3407 case OFPUTIL_OFPT_PACKET_OUT:
3408 return handle_packet_out(ofconn, msg->data);
3410 case OFPUTIL_OFPT_PORT_MOD:
3411 return handle_port_mod(ofconn, oh);
3413 case OFPUTIL_OFPT_FLOW_MOD:
3414 return handle_flow_mod(ofconn, oh);
3416 case OFPUTIL_OFPT_BARRIER_REQUEST:
3417 return handle_barrier_request(ofconn, oh);
3419 /* OpenFlow replies. */
3420 case OFPUTIL_OFPT_ECHO_REPLY:
3423 /* Nicira extension requests. */
3424 case OFPUTIL_NXT_ROLE_REQUEST:
3425 return handle_role_request(ofconn, oh);
3427 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3428 return handle_nxt_flow_mod_table_id(ofconn, oh);
3430 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3431 return handle_nxt_set_flow_format(ofconn, oh);
3433 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3434 return handle_nxt_set_packet_in_format(ofconn, oh);
3436 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3437 return handle_nxt_set_controller_id(ofconn, oh);
3439 case OFPUTIL_NXT_FLOW_MOD:
3440 return handle_flow_mod(ofconn, oh);
3442 case OFPUTIL_NXT_FLOW_AGE:
3443 /* Nothing to do. */
3446 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3447 return handle_nxt_set_async_config(ofconn, oh);
3449 /* Statistics requests. */
3450 case OFPUTIL_OFPST_DESC_REQUEST:
3451 return handle_desc_stats_request(ofconn, msg->data);
3453 case OFPUTIL_OFPST_FLOW_REQUEST:
3454 case OFPUTIL_NXST_FLOW_REQUEST:
3455 return handle_flow_stats_request(ofconn, msg->data);
3457 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3458 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3459 return handle_aggregate_stats_request(ofconn, msg->data);
3461 case OFPUTIL_OFPST_TABLE_REQUEST:
3462 return handle_table_stats_request(ofconn, msg->data);
3464 case OFPUTIL_OFPST_PORT_REQUEST:
3465 return handle_port_stats_request(ofconn, msg->data);
3467 case OFPUTIL_OFPST_QUEUE_REQUEST:
3468 return handle_queue_stats_request(ofconn, msg->data);
3470 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3471 return handle_port_desc_stats_request(ofconn, msg->data);
3473 case OFPUTIL_MSG_INVALID:
3474 case OFPUTIL_OFPT_HELLO:
3475 case OFPUTIL_OFPT_ERROR:
3476 case OFPUTIL_OFPT_FEATURES_REPLY:
3477 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3478 case OFPUTIL_OFPT_PACKET_IN:
3479 case OFPUTIL_OFPT_FLOW_REMOVED:
3480 case OFPUTIL_OFPT_PORT_STATUS:
3481 case OFPUTIL_OFPT_BARRIER_REPLY:
3482 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3483 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3484 case OFPUTIL_OFPST_DESC_REPLY:
3485 case OFPUTIL_OFPST_FLOW_REPLY:
3486 case OFPUTIL_OFPST_QUEUE_REPLY:
3487 case OFPUTIL_OFPST_PORT_REPLY:
3488 case OFPUTIL_OFPST_TABLE_REPLY:
3489 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3490 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3491 case OFPUTIL_NXT_ROLE_REPLY:
3492 case OFPUTIL_NXT_FLOW_REMOVED:
3493 case OFPUTIL_NXT_PACKET_IN:
3494 case OFPUTIL_NXST_FLOW_REPLY:
3495 case OFPUTIL_NXST_AGGREGATE_REPLY:
3497 return (oh->type == OFPT10_STATS_REQUEST ||
3498 oh->type == OFPT10_STATS_REPLY
3499 ? OFPERR_OFPBRC_BAD_STAT
3500 : OFPERR_OFPBRC_BAD_TYPE);
3505 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3507 int error = handle_openflow__(ofconn, ofp_msg);
3508 if (error && error != OFPROTO_POSTPONE) {
3509 ofconn_send_error(ofconn, ofp_msg->data, error);
3511 COVERAGE_INC(ofproto_recv_openflow);
3512 return error != OFPROTO_POSTPONE;
3515 /* Asynchronous operations. */
3517 /* Creates and returns a new ofopgroup that is not associated with any
3518 * OpenFlow connection.
3520 * The caller should add operations to the returned group with
3521 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3522 static struct ofopgroup *
3523 ofopgroup_create_unattached(struct ofproto *ofproto)
3525 struct ofopgroup *group = xzalloc(sizeof *group);
3526 group->ofproto = ofproto;
3527 list_init(&group->ofproto_node);
3528 list_init(&group->ops);
3529 list_init(&group->ofconn_node);
3533 /* Creates and returns a new ofopgroup for 'ofproto'.
3535 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3536 * connection. The 'request' and 'buffer_id' arguments are ignored.
3538 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3539 * If the ofopgroup eventually fails, then the error reply will include
3540 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3541 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3543 * The caller should add operations to the returned group with
3544 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3545 static struct ofopgroup *
3546 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3547 const struct ofp_header *request, uint32_t buffer_id)
3549 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3551 size_t request_len = ntohs(request->length);
3553 assert(ofconn_get_ofproto(ofconn) == ofproto);
3555 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3556 group->ofconn = ofconn;
3557 group->request = xmemdup(request, MIN(request_len, 64));
3558 group->buffer_id = buffer_id;
3563 /* Submits 'group' for processing.
3565 * If 'group' contains no operations (e.g. none were ever added, or all of the
3566 * ones that were added completed synchronously), then it is destroyed
3567 * immediately. Otherwise it is added to the ofproto's list of pending
3570 ofopgroup_submit(struct ofopgroup *group)
3572 if (list_is_empty(&group->ops)) {
3573 ofopgroup_destroy(group);
3575 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3576 group->ofproto->n_pending++;
3581 ofopgroup_destroy(struct ofopgroup *group)
3583 assert(list_is_empty(&group->ops));
3584 if (!list_is_empty(&group->ofproto_node)) {
3585 assert(group->ofproto->n_pending > 0);
3586 group->ofproto->n_pending--;
3587 list_remove(&group->ofproto_node);
3589 if (!list_is_empty(&group->ofconn_node)) {
3590 list_remove(&group->ofconn_node);
3592 ofconn_send_error(group->ofconn, group->request, group->error);
3594 connmgr_retry(group->ofproto->connmgr);
3596 free(group->request);
3600 /* Initiates a new operation on 'rule', of the specified 'type', within
3601 * 'group'. Prior to calling, 'rule' must not have any pending operation.
3603 * Returns the newly created ofoperation (which is also available as
3604 * rule->pending). */
3605 static struct ofoperation *
3606 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3607 enum ofoperation_type type)
3609 struct ofproto *ofproto = group->ofproto;
3610 struct ofoperation *op;
3612 assert(!rule->pending);
3614 op = rule->pending = xzalloc(sizeof *op);
3616 list_push_back(&group->ops, &op->group_node);
3619 op->flow_cookie = rule->flow_cookie;
3621 if (type == OFOPERATION_DELETE) {
3622 hmap_insert(&ofproto->deletions, &op->hmap_node,
3623 cls_rule_hash(&rule->cr, rule->table_id));
3630 ofoperation_destroy(struct ofoperation *op)
3632 struct ofopgroup *group = op->group;
3635 op->rule->pending = NULL;
3637 if (op->type == OFOPERATION_DELETE) {
3638 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3640 list_remove(&op->group_node);
3644 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3645 ofopgroup_destroy(group);
3649 /* Indicates that 'op' completed with status 'error', which is either 0 to
3650 * indicate success or an OpenFlow error code on failure.
3652 * If 'error' is 0, indicating success, the operation will be committed
3653 * permanently to the flow table. There is one interesting subcase:
3655 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3656 * the flow table (the "victim" rule) by a new one, then the caller must
3657 * have uninitialized any derived state in the victim rule, as in step 5 in
3658 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3659 * performs steps 6 and 7 for the victim rule, most notably by calling its
3660 * ->rule_dealloc() function.
3662 * If 'error' is nonzero, then generally the operation will be rolled back:
3664 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3665 * restores the original rule. The caller must have uninitialized any
3666 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3667 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3668 * and 7 for the new rule, calling its ->rule_dealloc() function.
3670 * - If 'op' is a "modify flow" operation, ofproto restores the original
3673 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3674 * allowed to fail. It must always succeed.
3676 * Please see the large comment in ofproto/ofproto-provider.h titled
3677 * "Asynchronous Operation Support" for more information. */
3679 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3681 struct ofopgroup *group = op->group;
3682 struct rule *rule = op->rule;
3683 struct ofproto *ofproto = rule->ofproto;
3685 assert(rule->pending == op);
3689 && op->type != OFOPERATION_DELETE
3691 && group->buffer_id != UINT32_MAX
3692 && list_is_singleton(&op->group_node)) {
3693 struct ofpbuf *packet;
3696 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3700 error = rule_execute(rule, in_port, packet);
3703 if (!group->error) {
3704 group->error = error;
3708 case OFOPERATION_ADD:
3710 ofproto_rule_destroy__(op->victim);
3711 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3712 == htons(VLAN_VID_MASK)) {
3713 if (ofproto->vlan_bitmap) {
3714 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3716 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3717 bitmap_set1(ofproto->vlan_bitmap, vid);
3718 ofproto->vlans_changed = true;
3721 ofproto->vlans_changed = true;
3725 oftable_substitute_rule(rule, op->victim);
3726 ofproto_rule_destroy__(rule);
3731 case OFOPERATION_DELETE:
3733 ofproto_rule_destroy__(rule);
3737 case OFOPERATION_MODIFY:
3739 rule->modified = time_msec();
3741 free(rule->ofpacts);
3742 rule->ofpacts = op->ofpacts;
3743 rule->ofpacts_len = op->ofpacts_len;
3751 ofoperation_destroy(op);
3755 ofoperation_get_victim(struct ofoperation *op)
3757 assert(op->type == OFOPERATION_ADD);
3762 pick_datapath_id(const struct ofproto *ofproto)
3764 const struct ofport *port;
3766 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3768 uint8_t ea[ETH_ADDR_LEN];
3771 error = netdev_get_etheraddr(port->netdev, ea);
3773 return eth_addr_to_uint64(ea);
3775 VLOG_WARN("%s: could not get MAC address for %s (%s)",
3776 ofproto->name, netdev_get_name(port->netdev),
3779 return ofproto->fallback_dpid;
3783 pick_fallback_dpid(void)
3785 uint8_t ea[ETH_ADDR_LEN];
3786 eth_addr_nicira_random(ea);
3787 return eth_addr_to_uint64(ea);
3790 /* Table overflow policy. */
3792 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3793 * is not configured to evict rules or if the table contains no evictable
3794 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3796 static struct rule *
3797 choose_rule_to_evict(struct oftable *table)
3799 struct eviction_group *evg;
3801 if (!table->eviction_fields) {
3805 /* In the common case, the outer and inner loops here will each be entered
3808 * - The inner loop normally "return"s in its first iteration. If the
3809 * eviction group has any evictable rules, then it always returns in
3812 * - The outer loop only iterates more than once if the largest eviction
3813 * group has no evictable rules.
3815 * - The outer loop can exit only if table's 'max_flows' is all filled up
3816 * by unevictable rules'. */
3817 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
3820 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
3821 if (rule->evictable) {
3830 /* Searches 'ofproto' for tables that have more flows than their configured
3831 * maximum and that have flow eviction enabled, and evicts as many flows as
3832 * necessary and currently feasible from them.
3834 * This triggers only when an OpenFlow table has N flows in it and then the
3835 * client configures a maximum number of flows less than N. */
3837 ofproto_evict(struct ofproto *ofproto)
3839 struct ofopgroup *group;
3840 struct oftable *table;
3842 group = ofopgroup_create_unattached(ofproto);
3843 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
3844 while (classifier_count(&table->cls) > table->max_flows
3845 && table->eviction_fields) {
3848 rule = choose_rule_to_evict(table);
3849 if (!rule || rule->pending) {
3853 ofoperation_create(group, rule, OFOPERATION_DELETE);
3854 oftable_remove_rule(rule);
3855 ofproto->ofproto_class->rule_destruct(rule);
3858 ofopgroup_submit(group);
3861 /* Eviction groups. */
3863 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3864 * rules. The priority contains low-order random bits to ensure that eviction
3865 * groups with the same number of rules are prioritized randomly. */
3867 eviction_group_priority(size_t n_rules)
3869 uint16_t size = MIN(UINT16_MAX, n_rules);
3870 return (size << 16) | random_uint16();
3873 /* Updates 'evg', an eviction_group within 'table', following a change that
3874 * adds or removes rules in 'evg'. */
3876 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
3878 heap_change(&table->eviction_groups_by_size, &evg->size_node,
3879 eviction_group_priority(heap_count(&evg->rules)));
3882 /* Destroys 'evg', an eviction_group within 'table':
3884 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3885 * rules themselves, just removes them from the eviction group.)
3887 * - Removes 'evg' from 'table'.
3891 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
3893 while (!heap_is_empty(&evg->rules)) {
3896 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
3897 rule->eviction_group = NULL;
3899 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
3900 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
3901 heap_destroy(&evg->rules);
3905 /* Removes 'rule' from its eviction group, if any. */
3907 eviction_group_remove_rule(struct rule *rule)
3909 if (rule->eviction_group) {
3910 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3911 struct eviction_group *evg = rule->eviction_group;
3913 rule->eviction_group = NULL;
3914 heap_remove(&evg->rules, &rule->evg_node);
3915 if (heap_is_empty(&evg->rules)) {
3916 eviction_group_destroy(table, evg);
3918 eviction_group_resized(table, evg);
3923 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3924 * returns the hash value. */
3926 eviction_group_hash_rule(struct rule *rule)
3928 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3929 const struct mf_subfield *sf;
3932 hash = table->eviction_group_id_basis;
3933 for (sf = table->eviction_fields;
3934 sf < &table->eviction_fields[table->n_eviction_fields];
3937 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
3938 union mf_value value;
3940 mf_get_value(sf->field, &rule->cr.flow, &value);
3942 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
3944 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
3945 unsigned int start = sf->ofs + sf->n_bits;
3946 bitwise_zero(&value, sf->field->n_bytes, start,
3947 sf->field->n_bytes * 8 - start);
3949 hash = hash_bytes(&value, sf->field->n_bytes, hash);
3951 hash = hash_int(hash, 0);
3958 /* Returns an eviction group within 'table' with the given 'id', creating one
3960 static struct eviction_group *
3961 eviction_group_find(struct oftable *table, uint32_t id)
3963 struct eviction_group *evg;
3965 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
3969 evg = xmalloc(sizeof *evg);
3970 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
3971 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
3972 eviction_group_priority(0));
3973 heap_init(&evg->rules);
3978 /* Returns an eviction priority for 'rule'. The return value should be
3979 * interpreted so that higher priorities make a rule more attractive candidates
3982 rule_eviction_priority(struct rule *rule)
3984 long long int hard_expiration;
3985 long long int idle_expiration;
3986 long long int expiration;
3987 uint32_t expiration_offset;
3989 /* Calculate time of expiration. */
3990 hard_expiration = (rule->hard_timeout
3991 ? rule->modified + rule->hard_timeout * 1000
3993 idle_expiration = (rule->idle_timeout
3994 ? rule->used + rule->idle_timeout * 1000
3996 expiration = MIN(hard_expiration, idle_expiration);
3997 if (expiration == LLONG_MAX) {
4001 /* Calculate the time of expiration as a number of (approximate) seconds
4002 * after program startup.
4004 * This should work OK for program runs that last UINT32_MAX seconds or
4005 * less. Therefore, please restart OVS at least once every 136 years. */
4006 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4008 /* Invert the expiration offset because we're using a max-heap. */
4009 return UINT32_MAX - expiration_offset;
4012 /* Adds 'rule' to an appropriate eviction group for its oftable's
4013 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4014 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4017 * The caller must ensure that 'rule' is not already in an eviction group. */
4019 eviction_group_add_rule(struct rule *rule)
4021 struct ofproto *ofproto = rule->ofproto;
4022 struct oftable *table = &ofproto->tables[rule->table_id];
4024 if (table->eviction_fields
4025 && (rule->hard_timeout || rule->idle_timeout)) {
4026 struct eviction_group *evg;
4028 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4030 rule->eviction_group = evg;
4031 heap_insert(&evg->rules, &rule->evg_node,
4032 rule_eviction_priority(rule));
4033 eviction_group_resized(table, evg);
4039 /* Initializes 'table'. */
4041 oftable_init(struct oftable *table)
4043 memset(table, 0, sizeof *table);
4044 classifier_init(&table->cls);
4045 table->max_flows = UINT_MAX;
4048 /* Destroys 'table', including its classifier and eviction groups.
4050 * The caller is responsible for freeing 'table' itself. */
4052 oftable_destroy(struct oftable *table)
4054 assert(classifier_is_empty(&table->cls));
4055 oftable_disable_eviction(table);
4056 classifier_destroy(&table->cls);
4060 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4061 * string, then 'table' will use its default name.
4063 * This only affects the name exposed for a table exposed through the OpenFlow
4064 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4066 oftable_set_name(struct oftable *table, const char *name)
4068 if (name && name[0]) {
4069 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4070 if (!table->name || strncmp(name, table->name, len)) {
4072 table->name = xmemdup0(name, len);
4080 /* oftables support a choice of two policies when adding a rule would cause the
4081 * number of flows in the table to exceed the configured maximum number: either
4082 * they can refuse to add the new flow or they can evict some existing flow.
4083 * This function configures the former policy on 'table'. */
4085 oftable_disable_eviction(struct oftable *table)
4087 if (table->eviction_fields) {
4088 struct eviction_group *evg, *next;
4090 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4091 &table->eviction_groups_by_id) {
4092 eviction_group_destroy(table, evg);
4094 hmap_destroy(&table->eviction_groups_by_id);
4095 heap_destroy(&table->eviction_groups_by_size);
4097 free(table->eviction_fields);
4098 table->eviction_fields = NULL;
4099 table->n_eviction_fields = 0;
4103 /* oftables support a choice of two policies when adding a rule would cause the
4104 * number of flows in the table to exceed the configured maximum number: either
4105 * they can refuse to add the new flow or they can evict some existing flow.
4106 * This function configures the latter policy on 'table', with fairness based
4107 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4108 * 'n_fields' as 0 disables fairness.) */
4110 oftable_enable_eviction(struct oftable *table,
4111 const struct mf_subfield *fields, size_t n_fields)
4113 struct cls_cursor cursor;
4116 if (table->eviction_fields
4117 && n_fields == table->n_eviction_fields
4119 || !memcmp(fields, table->eviction_fields,
4120 n_fields * sizeof *fields))) {
4125 oftable_disable_eviction(table);
4127 table->n_eviction_fields = n_fields;
4128 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4130 table->eviction_group_id_basis = random_uint32();
4131 hmap_init(&table->eviction_groups_by_id);
4132 heap_init(&table->eviction_groups_by_size);
4134 cls_cursor_init(&cursor, &table->cls, NULL);
4135 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4136 eviction_group_add_rule(rule);
4140 /* Removes 'rule' from the oftable that contains it. */
4142 oftable_remove_rule(struct rule *rule)
4144 struct ofproto *ofproto = rule->ofproto;
4145 struct oftable *table = &ofproto->tables[rule->table_id];
4147 classifier_remove(&table->cls, &rule->cr);
4148 eviction_group_remove_rule(rule);
4151 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4152 * oftable that has an identical cls_rule. Returns the rule that was removed,
4153 * if any, and otherwise NULL. */
4154 static struct rule *
4155 oftable_replace_rule(struct rule *rule)
4157 struct ofproto *ofproto = rule->ofproto;
4158 struct oftable *table = &ofproto->tables[rule->table_id];
4159 struct rule *victim;
4161 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4163 eviction_group_remove_rule(victim);
4165 eviction_group_add_rule(rule);
4169 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4171 oftable_substitute_rule(struct rule *old, struct rule *new)
4174 oftable_replace_rule(new);
4176 oftable_remove_rule(old);
4180 /* unixctl commands. */
4183 ofproto_lookup(const char *name)
4185 struct ofproto *ofproto;
4187 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4189 if (!strcmp(ofproto->name, name)) {
4197 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4198 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4200 struct ofproto *ofproto;
4204 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4205 ds_put_format(&results, "%s\n", ofproto->name);
4207 unixctl_command_reply(conn, ds_cstr(&results));
4208 ds_destroy(&results);
4212 ofproto_unixctl_init(void)
4214 static bool registered;
4220 unixctl_command_register("ofproto/list", "", 0, 0,
4221 ofproto_unixctl_list, NULL);
4224 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4226 * This is deprecated. It is only for compatibility with broken device drivers
4227 * in old versions of Linux that do not properly support VLANs when VLAN
4228 * devices are not used. When broken device drivers are no longer in
4229 * widespread use, we will delete these interfaces. */
4231 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4232 * (exactly) by an OpenFlow rule in 'ofproto'. */
4234 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4236 const struct oftable *oftable;
4238 free(ofproto->vlan_bitmap);
4239 ofproto->vlan_bitmap = bitmap_allocate(4096);
4240 ofproto->vlans_changed = false;
4242 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4243 const struct cls_table *table;
4245 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4246 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4247 == htons(VLAN_VID_MASK)) {
4248 const struct cls_rule *rule;
4250 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4251 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4252 bitmap_set1(vlan_bitmap, vid);
4253 bitmap_set1(ofproto->vlan_bitmap, vid);
4260 /* Returns true if new VLANs have come into use by the flow table since the
4261 * last call to ofproto_get_vlan_usage().
4263 * We don't track when old VLANs stop being used. */
4265 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4267 return ofproto->vlans_changed;
4270 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4271 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4272 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4273 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4274 * then the VLAN device is un-enslaved. */
4276 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4277 uint16_t realdev_ofp_port, int vid)
4279 struct ofport *ofport;
4282 assert(vlandev_ofp_port != realdev_ofp_port);
4284 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4286 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4287 ofproto->name, vlandev_ofp_port);
4291 if (!ofproto->ofproto_class->set_realdev) {
4292 if (!vlandev_ofp_port) {
4295 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4299 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4301 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4302 ofproto->name, vlandev_ofp_port,
4303 netdev_get_name(ofport->netdev), strerror(error));