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-errors.h"
36 #include "ofp-print.h"
39 #include "ofproto-provider.h"
40 #include "openflow/nicira-ext.h"
41 #include "openflow/openflow.h"
45 #include "poll-loop.h"
51 #include "unaligned.h"
55 VLOG_DEFINE_THIS_MODULE(ofproto);
57 COVERAGE_DEFINE(ofproto_error);
58 COVERAGE_DEFINE(ofproto_flush);
59 COVERAGE_DEFINE(ofproto_no_packet_in);
60 COVERAGE_DEFINE(ofproto_packet_out);
61 COVERAGE_DEFINE(ofproto_queue_req);
62 COVERAGE_DEFINE(ofproto_recv_openflow);
63 COVERAGE_DEFINE(ofproto_reinit_ports);
64 COVERAGE_DEFINE(ofproto_uninstallable);
65 COVERAGE_DEFINE(ofproto_update_port);
68 S_OPENFLOW, /* Processing OpenFlow commands. */
69 S_EVICT, /* Evicting flows from over-limit tables. */
70 S_FLUSH, /* Deleting all flow table rules. */
73 enum ofoperation_type {
79 /* A single OpenFlow request can execute any number of operations. The
80 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
81 * ofconn to which an error reply should be sent if necessary.
83 * ofproto initiates some operations internally. These operations are still
84 * assigned to groups but will not have an associated ofconn. */
86 struct ofproto *ofproto; /* Owning ofproto. */
87 struct list ofproto_node; /* In ofproto's "pending" list. */
88 struct list ops; /* List of "struct ofoperation"s. */
90 /* Data needed to send OpenFlow reply on failure or to send a buffered
93 * If list_is_empty(ofconn_node) then this ofopgroup never had an
94 * associated ofconn or its ofconn's connection dropped after it initiated
95 * the operation. In the latter case 'ofconn' is a wild pointer that
96 * refers to freed memory, so the 'ofconn' member must be used only if
97 * !list_is_empty(ofconn_node).
99 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
100 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
101 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
102 uint32_t buffer_id; /* Buffer id from original request. */
103 int error; /* 0 if no error yet, otherwise error code. */
106 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
107 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
108 const struct ofp_header *,
110 static void ofopgroup_submit(struct ofopgroup *);
111 static void ofopgroup_destroy(struct ofopgroup *);
113 /* A single flow table operation. */
115 struct ofopgroup *group; /* Owning group. */
116 struct list group_node; /* In ofopgroup's "ops" list. */
117 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
118 struct rule *rule; /* Rule being operated upon. */
119 enum ofoperation_type type; /* Type of operation. */
120 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
121 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
122 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
123 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
126 static void ofoperation_create(struct ofopgroup *, struct rule *,
127 enum ofoperation_type);
128 static void ofoperation_destroy(struct ofoperation *);
131 static void oftable_init(struct oftable *);
132 static void oftable_destroy(struct oftable *);
134 static void oftable_set_name(struct oftable *, const char *name);
136 static void oftable_disable_eviction(struct oftable *);
137 static void oftable_enable_eviction(struct oftable *,
138 const struct mf_subfield *fields,
141 static void oftable_remove_rule(struct rule *);
142 static struct rule *oftable_replace_rule(struct rule *);
143 static void oftable_substitute_rule(struct rule *old, struct rule *new);
145 /* A set of rules within a single OpenFlow table (oftable) that have the same
146 * values for the oftable's eviction_fields. A rule to be evicted, when one is
147 * needed, is taken from the eviction group that contains the greatest number
150 * An oftable owns any number of eviction groups, each of which contains any
153 * Membership in an eviction group is imprecise, based on the hash of the
154 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
155 * That is, if two rules have different eviction_fields, but those
156 * eviction_fields hash to the same value, then they will belong to the same
157 * eviction_group anyway.
159 * (When eviction is not enabled on an oftable, we don't track any eviction
160 * groups, to save time and space.) */
161 struct eviction_group {
162 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
163 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
164 struct heap rules; /* Contains "struct rule"s. */
167 static struct rule *choose_rule_to_evict(struct oftable *);
168 static void ofproto_evict(struct ofproto *);
169 static uint32_t rule_eviction_priority(struct rule *);
172 static void ofport_destroy__(struct ofport *);
173 static void ofport_destroy(struct ofport *);
175 static void update_port(struct ofproto *, const char *devname);
176 static int init_ports(struct ofproto *);
177 static void reinit_ports(struct ofproto *);
180 static void ofproto_rule_destroy__(struct rule *);
181 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
182 static bool rule_is_modifiable(const struct rule *);
183 static bool rule_is_hidden(const struct rule *);
186 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
187 const struct ofputil_flow_mod *,
188 const struct ofp_header *);
189 static void delete_flow__(struct rule *, struct ofopgroup *);
190 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
191 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
192 const struct ofputil_flow_mod *,
193 const struct ofp_header *);
196 static uint64_t pick_datapath_id(const struct ofproto *);
197 static uint64_t pick_fallback_dpid(void);
198 static void ofproto_destroy__(struct ofproto *);
199 static void update_mtu(struct ofproto *, struct ofport *);
202 static void ofproto_unixctl_init(void);
204 /* All registered ofproto classes, in probe order. */
205 static const struct ofproto_class **ofproto_classes;
206 static size_t n_ofproto_classes;
207 static size_t allocated_ofproto_classes;
209 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
210 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
212 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
215 ofproto_initialize(void)
221 ofproto_class_register(&ofproto_dpif_class);
225 /* 'type' should be a normalized datapath type, as returned by
226 * ofproto_normalize_type(). Returns the corresponding ofproto_class
227 * structure, or a null pointer if there is none registered for 'type'. */
228 static const struct ofproto_class *
229 ofproto_class_find__(const char *type)
233 ofproto_initialize();
234 for (i = 0; i < n_ofproto_classes; i++) {
235 const struct ofproto_class *class = ofproto_classes[i];
240 class->enumerate_types(&types);
241 found = sset_contains(&types, type);
242 sset_destroy(&types);
248 VLOG_WARN("unknown datapath type %s", type);
252 /* Registers a new ofproto class. After successful registration, new ofprotos
253 * of that type can be created using ofproto_create(). */
255 ofproto_class_register(const struct ofproto_class *new_class)
259 for (i = 0; i < n_ofproto_classes; i++) {
260 if (ofproto_classes[i] == new_class) {
265 if (n_ofproto_classes >= allocated_ofproto_classes) {
266 ofproto_classes = x2nrealloc(ofproto_classes,
267 &allocated_ofproto_classes,
268 sizeof *ofproto_classes);
270 ofproto_classes[n_ofproto_classes++] = new_class;
274 /* Unregisters a datapath provider. 'type' must have been previously
275 * registered and not currently be in use by any ofprotos. After
276 * unregistration new datapaths of that type cannot be opened using
277 * ofproto_create(). */
279 ofproto_class_unregister(const struct ofproto_class *class)
283 for (i = 0; i < n_ofproto_classes; i++) {
284 if (ofproto_classes[i] == class) {
285 for (i++; i < n_ofproto_classes; i++) {
286 ofproto_classes[i - 1] = ofproto_classes[i];
292 VLOG_WARN("attempted to unregister an ofproto class that is not "
297 /* Clears 'types' and enumerates all registered ofproto types into it. The
298 * caller must first initialize the sset. */
300 ofproto_enumerate_types(struct sset *types)
304 ofproto_initialize();
305 for (i = 0; i < n_ofproto_classes; i++) {
306 ofproto_classes[i]->enumerate_types(types);
310 /* Returns the fully spelled out name for the given ofproto 'type'.
312 * Normalized type string can be compared with strcmp(). Unnormalized type
313 * string might be the same even if they have different spellings. */
315 ofproto_normalize_type(const char *type)
317 return type && type[0] ? type : "system";
320 /* Clears 'names' and enumerates the names of all known created ofprotos with
321 * the given 'type'. The caller must first initialize the sset. Returns 0 if
322 * successful, otherwise a positive errno value.
324 * Some kinds of datapaths might not be practically enumerable. This is not
325 * considered an error. */
327 ofproto_enumerate_names(const char *type, struct sset *names)
329 const struct ofproto_class *class = ofproto_class_find__(type);
330 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
334 ofproto_create(const char *datapath_name, const char *datapath_type,
335 struct ofproto **ofprotop)
337 const struct ofproto_class *class;
338 struct ofproto *ofproto;
343 ofproto_initialize();
344 ofproto_unixctl_init();
346 datapath_type = ofproto_normalize_type(datapath_type);
347 class = ofproto_class_find__(datapath_type);
349 VLOG_WARN("could not create datapath %s of unknown type %s",
350 datapath_name, datapath_type);
354 ofproto = class->alloc();
356 VLOG_ERR("failed to allocate datapath %s of type %s",
357 datapath_name, datapath_type);
362 memset(ofproto, 0, sizeof *ofproto);
363 ofproto->ofproto_class = class;
364 ofproto->name = xstrdup(datapath_name);
365 ofproto->type = xstrdup(datapath_type);
366 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
367 hash_string(ofproto->name, 0));
368 ofproto->datapath_id = 0;
369 ofproto_set_flow_eviction_threshold(ofproto,
370 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
371 ofproto->forward_bpdu = false;
372 ofproto->fallback_dpid = pick_fallback_dpid();
373 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
374 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
375 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
376 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
377 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
378 ofproto->frag_handling = OFPC_FRAG_NORMAL;
379 hmap_init(&ofproto->ports);
380 shash_init(&ofproto->port_by_name);
381 ofproto->tables = NULL;
382 ofproto->n_tables = 0;
383 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
384 ofproto->state = S_OPENFLOW;
385 list_init(&ofproto->pending);
386 ofproto->n_pending = 0;
387 hmap_init(&ofproto->deletions);
388 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
389 ofproto->first_op = ofproto->last_op = LLONG_MIN;
390 ofproto->next_op_report = LLONG_MAX;
391 ofproto->op_backoff = LLONG_MIN;
392 ofproto->vlan_bitmap = NULL;
393 ofproto->vlans_changed = false;
394 ofproto->min_mtu = INT_MAX;
396 error = ofproto->ofproto_class->construct(ofproto);
398 VLOG_ERR("failed to open datapath %s: %s",
399 datapath_name, strerror(error));
400 ofproto_destroy__(ofproto);
404 assert(ofproto->n_tables);
406 ofproto->datapath_id = pick_datapath_id(ofproto);
407 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
415 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
417 struct oftable *table;
419 assert(!ofproto->n_tables);
420 assert(n_tables >= 1 && n_tables <= 255);
422 ofproto->n_tables = n_tables;
423 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
424 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
430 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
432 uint64_t old_dpid = p->datapath_id;
433 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
434 if (p->datapath_id != old_dpid) {
435 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
437 /* Force all active connections to reconnect, since there is no way to
438 * notify a controller that the datapath ID has changed. */
439 ofproto_reconnect_controllers(p);
444 ofproto_set_controllers(struct ofproto *p,
445 const struct ofproto_controller *controllers,
446 size_t n_controllers)
448 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
452 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
454 connmgr_set_fail_mode(p->connmgr, fail_mode);
457 /* Drops the connections between 'ofproto' and all of its controllers, forcing
458 * them to reconnect. */
460 ofproto_reconnect_controllers(struct ofproto *ofproto)
462 connmgr_reconnect(ofproto->connmgr);
465 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
466 * in-band control should guarantee access, in the same way that in-band
467 * control guarantees access to OpenFlow controllers. */
469 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
470 const struct sockaddr_in *extras, size_t n)
472 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
475 /* Sets the OpenFlow queue used by flows set up by in-band control on
476 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
477 * flows will use the default queue. */
479 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
481 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
484 /* Sets the number of flows at which eviction from the kernel flow table
487 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
489 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
490 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
492 ofproto->flow_eviction_threshold = threshold;
496 /* If forward_bpdu is true, the NORMAL action will forward frames with
497 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
498 * the NORMAL action will drop these frames. */
500 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
502 bool old_val = ofproto->forward_bpdu;
503 ofproto->forward_bpdu = forward_bpdu;
504 if (old_val != ofproto->forward_bpdu) {
505 if (ofproto->ofproto_class->forward_bpdu_changed) {
506 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
511 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
512 * 'idle_time', in seconds. */
514 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
516 if (ofproto->ofproto_class->set_mac_idle_time) {
517 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
522 ofproto_set_desc(struct ofproto *p,
523 const char *mfr_desc, const char *hw_desc,
524 const char *sw_desc, const char *serial_desc,
527 struct ofp_desc_stats *ods;
530 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
531 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
532 sizeof ods->mfr_desc);
535 p->mfr_desc = xstrdup(mfr_desc);
538 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
539 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
540 sizeof ods->hw_desc);
543 p->hw_desc = xstrdup(hw_desc);
546 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
547 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
548 sizeof ods->sw_desc);
551 p->sw_desc = xstrdup(sw_desc);
554 if (strlen(serial_desc) >= sizeof ods->serial_num) {
555 VLOG_WARN("truncating serial_desc, must be less than %zu "
557 sizeof ods->serial_num);
559 free(p->serial_desc);
560 p->serial_desc = xstrdup(serial_desc);
563 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
564 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
565 sizeof ods->dp_desc);
568 p->dp_desc = xstrdup(dp_desc);
573 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
575 return connmgr_set_snoops(ofproto->connmgr, snoops);
579 ofproto_set_netflow(struct ofproto *ofproto,
580 const struct netflow_options *nf_options)
582 if (nf_options && sset_is_empty(&nf_options->collectors)) {
586 if (ofproto->ofproto_class->set_netflow) {
587 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
589 return nf_options ? EOPNOTSUPP : 0;
594 ofproto_set_sflow(struct ofproto *ofproto,
595 const struct ofproto_sflow_options *oso)
597 if (oso && sset_is_empty(&oso->targets)) {
601 if (ofproto->ofproto_class->set_sflow) {
602 return ofproto->ofproto_class->set_sflow(ofproto, oso);
604 return oso ? EOPNOTSUPP : 0;
608 /* Spanning Tree Protocol (STP) configuration. */
610 /* Configures STP on 'ofproto' using the settings defined in 's'. If
611 * 's' is NULL, disables STP.
613 * Returns 0 if successful, otherwise a positive errno value. */
615 ofproto_set_stp(struct ofproto *ofproto,
616 const struct ofproto_stp_settings *s)
618 return (ofproto->ofproto_class->set_stp
619 ? ofproto->ofproto_class->set_stp(ofproto, s)
623 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
624 * 'enabled' member of 's' is false, then the other members are not
627 * Returns 0 if successful, otherwise a positive errno value. */
629 ofproto_get_stp_status(struct ofproto *ofproto,
630 struct ofproto_stp_status *s)
632 return (ofproto->ofproto_class->get_stp_status
633 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
637 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
638 * in 's'. The caller is responsible for assigning STP port numbers
639 * (using the 'port_num' member in the range of 1 through 255, inclusive)
640 * and ensuring there are no duplicates. If the 's' is NULL, then STP
641 * is disabled on the port.
643 * Returns 0 if successful, otherwise a positive errno value.*/
645 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
646 const struct ofproto_port_stp_settings *s)
648 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
650 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
651 ofproto->name, ofp_port);
655 return (ofproto->ofproto_class->set_stp_port
656 ? ofproto->ofproto_class->set_stp_port(ofport, s)
660 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
661 * 's'. If the 'enabled' member in 's' is false, then the other members
662 * are not meaningful.
664 * Returns 0 if successful, otherwise a positive errno value.*/
666 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
667 struct ofproto_port_stp_status *s)
669 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
671 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
672 "port %"PRIu16, ofproto->name, ofp_port);
676 return (ofproto->ofproto_class->get_stp_port_status
677 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
681 /* Queue DSCP configuration. */
683 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
684 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
685 * to implement QoS. Instead, it is used to implement features which require
686 * knowledge of what queues exist on a port, and some basic information about
689 * Returns 0 if successful, otherwise a positive errno value. */
691 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
692 const struct ofproto_port_queue *queues,
695 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
698 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
699 ofproto->name, ofp_port);
703 return (ofproto->ofproto_class->set_queues
704 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
708 /* Connectivity Fault Management configuration. */
710 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
712 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
714 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
715 if (ofport && ofproto->ofproto_class->set_cfm) {
716 ofproto->ofproto_class->set_cfm(ofport, NULL);
720 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
721 * basic configuration from the configuration members in 'cfm', and the remote
722 * maintenance point ID from remote_mpid. Ignores the statistics members of
725 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
727 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
728 const struct cfm_settings *s)
730 struct ofport *ofport;
733 ofport = ofproto_get_port(ofproto, ofp_port);
735 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
736 ofproto->name, ofp_port);
740 /* XXX: For configuration simplicity, we only support one remote_mpid
741 * outside of the CFM module. It's not clear if this is the correct long
742 * term solution or not. */
743 error = (ofproto->ofproto_class->set_cfm
744 ? ofproto->ofproto_class->set_cfm(ofport, s)
747 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
748 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
753 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
754 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
755 * 0 if LACP partner information is not current (generally indicating a
756 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
758 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
760 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
761 return (ofport && ofproto->ofproto_class->port_is_lacp_current
762 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
768 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
769 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
770 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
771 * configuration plus, if there is more than one slave, a bonding
774 * If 'aux' is already registered then this function updates its configuration
775 * to 's'. Otherwise, this function registers a new bundle.
777 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
780 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
781 const struct ofproto_bundle_settings *s)
783 return (ofproto->ofproto_class->bundle_set
784 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
788 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
789 * If no such bundle has been registered, this has no effect. */
791 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
793 return ofproto_bundle_register(ofproto, aux, NULL);
797 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
798 * If 'aux' is already registered then this function updates its configuration
799 * to 's'. Otherwise, this function registers a new mirror. */
801 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
802 const struct ofproto_mirror_settings *s)
804 return (ofproto->ofproto_class->mirror_set
805 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
809 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
810 * If no mirror has been registered, this has no effect. */
812 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
814 return ofproto_mirror_register(ofproto, aux, NULL);
817 /* Retrieves statistics from mirror associated with client data pointer
818 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
819 * 'bytes', respectively. If a particular counters is not supported,
820 * the appropriate argument is set to UINT64_MAX. */
822 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
823 uint64_t *packets, uint64_t *bytes)
825 if (!ofproto->ofproto_class->mirror_get_stats) {
826 *packets = *bytes = UINT64_MAX;
830 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
834 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
835 * which all packets are flooded, instead of using MAC learning. If
836 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
838 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
841 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
843 return (ofproto->ofproto_class->set_flood_vlans
844 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
848 /* Returns true if 'aux' is a registered bundle that is currently in use as the
849 * output for a mirror. */
851 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
853 return (ofproto->ofproto_class->is_mirror_output_bundle
854 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
858 /* Configuration of OpenFlow tables. */
860 /* Returns the number of OpenFlow tables in 'ofproto'. */
862 ofproto_get_n_tables(const struct ofproto *ofproto)
864 return ofproto->n_tables;
867 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
868 * settings from 's'. 'table_id' must be in the range 0 through the number of
869 * OpenFlow tables in 'ofproto' minus 1, inclusive.
871 * For read-only tables, only the name may be configured. */
873 ofproto_configure_table(struct ofproto *ofproto, int table_id,
874 const struct ofproto_table_settings *s)
876 struct oftable *table;
878 assert(table_id >= 0 && table_id < ofproto->n_tables);
879 table = &ofproto->tables[table_id];
881 oftable_set_name(table, s->name);
883 if (table->flags & OFTABLE_READONLY) {
888 oftable_enable_eviction(table, s->groups, s->n_groups);
890 oftable_disable_eviction(table);
893 table->max_flows = s->max_flows;
894 if (classifier_count(&table->cls) > table->max_flows
895 && table->eviction_fields) {
896 /* 'table' contains more flows than allowed. We might not be able to
897 * evict them right away because of the asynchronous nature of flow
898 * table changes. Schedule eviction for later. */
899 switch (ofproto->state) {
901 ofproto->state = S_EVICT;
905 /* We're already deleting flows, nothing more to do. */
912 ofproto_has_snoops(const struct ofproto *ofproto)
914 return connmgr_has_snoops(ofproto->connmgr);
918 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
920 connmgr_get_snoops(ofproto->connmgr, snoops);
924 ofproto_flush__(struct ofproto *ofproto)
926 struct ofopgroup *group;
927 struct oftable *table;
929 if (ofproto->ofproto_class->flush) {
930 ofproto->ofproto_class->flush(ofproto);
933 group = ofopgroup_create_unattached(ofproto);
934 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
935 struct rule *rule, *next_rule;
936 struct cls_cursor cursor;
938 if (table->flags & OFTABLE_HIDDEN) {
942 cls_cursor_init(&cursor, &table->cls, NULL);
943 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
944 if (!rule->pending) {
945 ofoperation_create(group, rule, OFOPERATION_DELETE);
946 oftable_remove_rule(rule);
947 ofproto->ofproto_class->rule_destruct(rule);
951 ofopgroup_submit(group);
955 ofproto_destroy__(struct ofproto *ofproto)
957 struct oftable *table;
959 assert(list_is_empty(&ofproto->pending));
960 assert(!ofproto->n_pending);
962 connmgr_destroy(ofproto->connmgr);
964 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
967 free(ofproto->mfr_desc);
968 free(ofproto->hw_desc);
969 free(ofproto->sw_desc);
970 free(ofproto->serial_desc);
971 free(ofproto->dp_desc);
972 hmap_destroy(&ofproto->ports);
973 shash_destroy(&ofproto->port_by_name);
975 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
976 oftable_destroy(table);
978 free(ofproto->tables);
980 hmap_destroy(&ofproto->deletions);
982 free(ofproto->vlan_bitmap);
984 ofproto->ofproto_class->dealloc(ofproto);
988 ofproto_destroy(struct ofproto *p)
990 struct ofport *ofport, *next_ofport;
997 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
998 ofport_destroy(ofport);
1001 p->ofproto_class->destruct(p);
1002 ofproto_destroy__(p);
1005 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1006 * kernel datapath, for example, this destroys the datapath in the kernel, and
1007 * with the netdev-based datapath, it tears down the data structures that
1008 * represent the datapath.
1010 * The datapath should not be currently open as an ofproto. */
1012 ofproto_delete(const char *name, const char *type)
1014 const struct ofproto_class *class = ofproto_class_find__(type);
1015 return (!class ? EAFNOSUPPORT
1016 : !class->del ? EACCES
1017 : class->del(type, name));
1021 process_port_change(struct ofproto *ofproto, int error, char *devname)
1023 if (error == ENOBUFS) {
1024 reinit_ports(ofproto);
1025 } else if (!error) {
1026 update_port(ofproto, devname);
1032 ofproto_run(struct ofproto *p)
1034 struct sset changed_netdevs;
1035 const char *changed_netdev;
1036 struct ofport *ofport;
1039 error = p->ofproto_class->run(p);
1040 if (error && error != EAGAIN) {
1041 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1044 if (p->ofproto_class->port_poll) {
1047 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1048 process_port_change(p, error, devname);
1052 /* Update OpenFlow port status for any port whose netdev has changed.
1054 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1055 * destroyed, so it's not safe to update ports directly from the
1056 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1057 * need this two-phase approach. */
1058 sset_init(&changed_netdevs);
1059 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1060 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1061 if (ofport->change_seq != change_seq) {
1062 ofport->change_seq = change_seq;
1063 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1066 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1067 update_port(p, changed_netdev);
1069 sset_destroy(&changed_netdevs);
1073 connmgr_run(p->connmgr, handle_openflow);
1077 connmgr_run(p->connmgr, NULL);
1079 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1080 p->state = S_OPENFLOW;
1085 connmgr_run(p->connmgr, NULL);
1087 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1088 connmgr_flushed(p->connmgr);
1089 p->state = S_OPENFLOW;
1097 if (time_msec() >= p->next_op_report) {
1098 long long int ago = (time_msec() - p->first_op) / 1000;
1099 long long int interval = (p->last_op - p->first_op) / 1000;
1103 ds_put_format(&s, "%d flow_mods ",
1104 p->n_add + p->n_delete + p->n_modify);
1105 if (interval == ago) {
1106 ds_put_format(&s, "in the last %lld s", ago);
1107 } else if (interval) {
1108 ds_put_format(&s, "in the %lld s starting %lld s ago",
1111 ds_put_format(&s, "%lld s ago", ago);
1114 ds_put_cstr(&s, " (");
1116 ds_put_format(&s, "%d adds, ", p->n_add);
1119 ds_put_format(&s, "%d deletes, ", p->n_delete);
1122 ds_put_format(&s, "%d modifications, ", p->n_modify);
1125 ds_put_char(&s, ')');
1127 VLOG_INFO("%s", ds_cstr(&s));
1130 p->n_add = p->n_delete = p->n_modify = 0;
1131 p->next_op_report = LLONG_MAX;
1137 /* Performs periodic activity required by 'ofproto' that needs to be done
1138 * with the least possible latency.
1140 * It makes sense to call this function a couple of times per poll loop, to
1141 * provide a significant performance boost on some benchmarks with the
1142 * ofproto-dpif implementation. */
1144 ofproto_run_fast(struct ofproto *p)
1148 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1149 if (error && error != EAGAIN) {
1150 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1151 p->name, strerror(error));
1157 ofproto_wait(struct ofproto *p)
1159 struct ofport *ofport;
1161 p->ofproto_class->wait(p);
1162 if (p->ofproto_class->port_poll_wait) {
1163 p->ofproto_class->port_poll_wait(p);
1166 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1167 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1168 poll_immediate_wake();
1174 connmgr_wait(p->connmgr, true);
1179 connmgr_wait(p->connmgr, false);
1180 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1181 poll_immediate_wake();
1188 ofproto_is_alive(const struct ofproto *p)
1190 return connmgr_has_controllers(p->connmgr);
1193 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1194 * memory_report(). */
1196 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1198 const struct oftable *table;
1199 unsigned int n_rules;
1201 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1202 simap_increase(usage, "ops",
1203 ofproto->n_pending + hmap_count(&ofproto->deletions));
1206 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1207 n_rules += classifier_count(&table->cls);
1209 simap_increase(usage, "rules", n_rules);
1211 if (ofproto->ofproto_class->get_memory_usage) {
1212 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1215 connmgr_get_memory_usage(ofproto->connmgr, usage);
1219 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1222 connmgr_get_controller_info(ofproto->connmgr, info);
1226 ofproto_free_ofproto_controller_info(struct shash *info)
1228 connmgr_free_controller_info(info);
1231 /* Makes a deep copy of 'old' into 'port'. */
1233 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1235 port->name = xstrdup(old->name);
1236 port->type = xstrdup(old->type);
1237 port->ofp_port = old->ofp_port;
1240 /* Frees memory allocated to members of 'ofproto_port'.
1242 * Do not call this function on an ofproto_port obtained from
1243 * ofproto_port_dump_next(): that function retains ownership of the data in the
1246 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1248 free(ofproto_port->name);
1249 free(ofproto_port->type);
1252 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1254 * This function provides no status indication. An error status for the entire
1255 * dump operation is provided when it is completed by calling
1256 * ofproto_port_dump_done().
1259 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1260 const struct ofproto *ofproto)
1262 dump->ofproto = ofproto;
1263 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1267 /* Attempts to retrieve another port from 'dump', which must have been created
1268 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1269 * 'port' and returns true. On failure, returns false.
1271 * Failure might indicate an actual error or merely that the last port has been
1272 * dumped. An error status for the entire dump operation is provided when it
1273 * is completed by calling ofproto_port_dump_done().
1275 * The ofproto owns the data stored in 'port'. It will remain valid until at
1276 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1277 * ofproto_port_dump_done(). */
1279 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1280 struct ofproto_port *port)
1282 const struct ofproto *ofproto = dump->ofproto;
1288 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1291 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1297 /* Completes port table dump operation 'dump', which must have been created
1298 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1299 * error-free, otherwise a positive errno value describing the problem. */
1301 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1303 const struct ofproto *ofproto = dump->ofproto;
1305 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1308 return dump->error == EOF ? 0 : dump->error;
1311 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1312 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1313 * is non-null). On failure, returns a positive errno value and sets
1314 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1316 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1317 uint16_t *ofp_portp)
1322 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1324 update_port(ofproto, netdev_get_name(netdev));
1327 *ofp_portp = error ? OFPP_NONE : ofp_port;
1332 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1333 * initializes '*port' appropriately; on failure, returns a positive errno
1336 * The caller owns the data in 'ofproto_port' and must free it with
1337 * ofproto_port_destroy() when it is no longer needed. */
1339 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1340 struct ofproto_port *port)
1344 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1346 memset(port, 0, sizeof *port);
1351 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1352 * Returns 0 if successful, otherwise a positive errno. */
1354 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1356 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1357 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1360 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1361 if (!error && ofport) {
1362 /* 'name' is the netdev's name and update_port() is going to close the
1363 * netdev. Just in case update_port() refers to 'name' after it
1364 * destroys 'ofport', make a copy of it around the update_port()
1366 char *devname = xstrdup(name);
1367 update_port(ofproto, devname);
1373 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1374 * performs the 'n_actions' actions in 'actions'. The new flow will not
1377 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1378 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1379 * controllers; otherwise, it will be hidden.
1381 * The caller retains ownership of 'cls_rule' and 'actions'.
1383 * This is a helper function for in-band control and fail-open. */
1385 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1386 const union ofp_action *actions, size_t n_actions)
1388 const struct rule *rule;
1390 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1391 &ofproto->tables[0].cls, cls_rule));
1392 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1393 actions, n_actions)) {
1394 struct ofputil_flow_mod fm;
1396 memset(&fm, 0, sizeof fm);
1398 fm.buffer_id = UINT32_MAX;
1399 fm.actions = (union ofp_action *) actions;
1400 fm.n_actions = n_actions;
1401 add_flow(ofproto, NULL, &fm, NULL);
1405 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1406 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1407 * operation cannot be initiated now but may be retried later.
1409 * This is a helper function for in-band control and fail-open. */
1411 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1413 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1416 /* Searches for a rule with matching criteria exactly equal to 'target' in
1417 * ofproto's table 0 and, if it finds one, deletes it.
1419 * This is a helper function for in-band control and fail-open. */
1421 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1425 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1426 &ofproto->tables[0].cls, target));
1428 /* No such rule -> success. */
1430 } else if (rule->pending) {
1431 /* An operation on the rule is already pending -> failure.
1432 * Caller must retry later if it's important. */
1435 /* Initiate deletion -> success. */
1436 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1437 ofoperation_create(group, rule, OFOPERATION_DELETE);
1438 oftable_remove_rule(rule);
1439 ofproto->ofproto_class->rule_destruct(rule);
1440 ofopgroup_submit(group);
1446 /* Starts the process of deleting all of the flows from all of ofproto's flow
1447 * tables and then reintroducing the flows required by in-band control and
1448 * fail-open. The process will complete in a later call to ofproto_run(). */
1450 ofproto_flush_flows(struct ofproto *ofproto)
1452 COVERAGE_INC(ofproto_flush);
1453 ofproto->state = S_FLUSH;
1457 reinit_ports(struct ofproto *p)
1459 struct ofproto_port_dump dump;
1460 struct sset devnames;
1461 struct ofport *ofport;
1462 struct ofproto_port ofproto_port;
1463 const char *devname;
1465 COVERAGE_INC(ofproto_reinit_ports);
1467 sset_init(&devnames);
1468 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1469 sset_add(&devnames, netdev_get_name(ofport->netdev));
1471 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1472 sset_add(&devnames, ofproto_port.name);
1475 SSET_FOR_EACH (devname, &devnames) {
1476 update_port(p, devname);
1478 sset_destroy(&devnames);
1481 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1482 * netdev cannot be opened. On success, also fills in 'opp'. */
1483 static struct netdev *
1484 ofport_open(const struct ofproto_port *ofproto_port,
1485 struct ofputil_phy_port *pp)
1487 enum netdev_flags flags;
1488 struct netdev *netdev;
1491 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1493 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1494 "cannot be opened (%s)",
1495 ofproto_port->name, ofproto_port->ofp_port,
1496 ofproto_port->name, strerror(error));
1500 pp->port_no = ofproto_port->ofp_port;
1501 netdev_get_etheraddr(netdev, pp->hw_addr);
1502 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1503 netdev_get_flags(netdev, &flags);
1504 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1505 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1506 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1507 &pp->supported, &pp->peer);
1508 pp->curr_speed = netdev_features_to_bps(pp->curr);
1509 pp->max_speed = netdev_features_to_bps(pp->supported);
1514 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1515 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1518 ofport_equal(const struct ofputil_phy_port *a,
1519 const struct ofputil_phy_port *b)
1521 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1522 && a->state == b->state
1523 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1524 && a->curr == b->curr
1525 && a->advertised == b->advertised
1526 && a->supported == b->supported
1527 && a->peer == b->peer
1528 && a->curr_speed == b->curr_speed
1529 && a->max_speed == b->max_speed);
1532 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1533 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1534 * one with the same name or port number). */
1536 ofport_install(struct ofproto *p,
1537 struct netdev *netdev, const struct ofputil_phy_port *pp)
1539 const char *netdev_name = netdev_get_name(netdev);
1540 struct ofport *ofport;
1543 /* Create ofport. */
1544 ofport = p->ofproto_class->port_alloc();
1549 ofport->ofproto = p;
1550 ofport->netdev = netdev;
1551 ofport->change_seq = netdev_change_seq(netdev);
1553 ofport->ofp_port = pp->port_no;
1555 /* Add port to 'p'. */
1556 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1557 shash_add(&p->port_by_name, netdev_name, ofport);
1559 update_mtu(p, ofport);
1561 /* Let the ofproto_class initialize its private data. */
1562 error = p->ofproto_class->port_construct(ofport);
1566 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1570 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1571 p->name, netdev_name, strerror(error));
1573 ofport_destroy__(ofport);
1575 netdev_close(netdev);
1579 /* Removes 'ofport' from 'p' and destroys it. */
1581 ofport_remove(struct ofport *ofport)
1583 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1585 ofport_destroy(ofport);
1588 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1591 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1593 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1595 ofport_remove(port);
1599 /* Updates 'port' with new 'pp' description.
1601 * Does not handle a name or port number change. The caller must implement
1602 * such a change as a delete followed by an add. */
1604 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1606 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1607 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1608 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1609 port->pp.state = pp->state;
1610 port->pp.curr = pp->curr;
1611 port->pp.advertised = pp->advertised;
1612 port->pp.supported = pp->supported;
1613 port->pp.peer = pp->peer;
1614 port->pp.curr_speed = pp->curr_speed;
1615 port->pp.max_speed = pp->max_speed;
1617 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1620 /* Update OpenFlow 'state' in 'port' and notify controller. */
1622 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1624 if (port->pp.state != state) {
1625 port->pp.state = state;
1626 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1632 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1634 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1636 if (port->ofproto->ofproto_class->set_realdev) {
1637 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1639 if (port->ofproto->ofproto_class->set_stp_port) {
1640 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1642 if (port->ofproto->ofproto_class->set_cfm) {
1643 port->ofproto->ofproto_class->set_cfm(port, NULL);
1645 if (port->ofproto->ofproto_class->bundle_remove) {
1646 port->ofproto->ofproto_class->bundle_remove(port);
1652 ofport_destroy__(struct ofport *port)
1654 struct ofproto *ofproto = port->ofproto;
1655 const char *name = netdev_get_name(port->netdev);
1657 hmap_remove(&ofproto->ports, &port->hmap_node);
1658 shash_delete(&ofproto->port_by_name,
1659 shash_find(&ofproto->port_by_name, name));
1661 netdev_close(port->netdev);
1662 ofproto->ofproto_class->port_dealloc(port);
1666 ofport_destroy(struct ofport *port)
1669 port->ofproto->ofproto_class->port_destruct(port);
1670 ofport_destroy__(port);
1675 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1677 struct ofport *port;
1679 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1680 hash_int(ofp_port, 0), &ofproto->ports) {
1681 if (port->ofp_port == ofp_port) {
1689 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1691 struct ofproto *ofproto = port->ofproto;
1694 if (ofproto->ofproto_class->port_get_stats) {
1695 error = ofproto->ofproto_class->port_get_stats(port, stats);
1704 update_port(struct ofproto *ofproto, const char *name)
1706 struct ofproto_port ofproto_port;
1707 struct ofputil_phy_port pp;
1708 struct netdev *netdev;
1709 struct ofport *port;
1711 COVERAGE_INC(ofproto_update_port);
1713 /* Fetch 'name''s location and properties from the datapath. */
1714 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1715 ? ofport_open(&ofproto_port, &pp)
1718 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1719 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1720 struct netdev *old_netdev = port->netdev;
1722 /* 'name' hasn't changed location. Any properties changed? */
1723 if (!ofport_equal(&port->pp, &pp)) {
1724 ofport_modified(port, &pp);
1727 update_mtu(ofproto, port);
1729 /* Install the newly opened netdev in case it has changed.
1730 * Don't close the old netdev yet in case port_modified has to
1731 * remove a retained reference to it.*/
1732 port->netdev = netdev;
1733 port->change_seq = netdev_change_seq(netdev);
1735 if (port->ofproto->ofproto_class->port_modified) {
1736 port->ofproto->ofproto_class->port_modified(port);
1739 netdev_close(old_netdev);
1741 /* If 'port' is nonnull then its name differs from 'name' and thus
1742 * we should delete it. If we think there's a port named 'name'
1743 * then its port number must be wrong now so delete it too. */
1745 ofport_remove(port);
1747 ofport_remove_with_name(ofproto, name);
1748 ofport_install(ofproto, netdev, &pp);
1751 /* Any port named 'name' is gone now. */
1752 ofport_remove_with_name(ofproto, name);
1754 ofproto_port_destroy(&ofproto_port);
1758 init_ports(struct ofproto *p)
1760 struct ofproto_port_dump dump;
1761 struct ofproto_port ofproto_port;
1763 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1764 uint16_t ofp_port = ofproto_port.ofp_port;
1765 if (ofproto_get_port(p, ofp_port)) {
1766 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1768 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1769 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1772 struct ofputil_phy_port pp;
1773 struct netdev *netdev;
1775 netdev = ofport_open(&ofproto_port, &pp);
1777 ofport_install(p, netdev, &pp);
1785 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1786 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1788 find_min_mtu(struct ofproto *p)
1790 struct ofport *ofport;
1793 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1794 struct netdev *netdev = ofport->netdev;
1797 /* Skip any internal ports, since that's what we're trying to
1799 if (!strcmp(netdev_get_type(netdev), "internal")) {
1803 if (netdev_get_mtu(netdev, &dev_mtu)) {
1806 if (!mtu || dev_mtu < mtu) {
1811 return mtu ? mtu: ETH_PAYLOAD_MAX;
1814 /* Update MTU of all datapath devices on 'p' to the minimum of the
1815 * non-datapath ports in event of 'port' added or changed. */
1817 update_mtu(struct ofproto *p, struct ofport *port)
1819 struct ofport *ofport;
1820 struct netdev *netdev = port->netdev;
1821 int dev_mtu, old_min;
1823 if (netdev_get_mtu(netdev, &dev_mtu)) {
1827 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1828 if (dev_mtu > p->min_mtu) {
1829 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1830 dev_mtu = p->min_mtu;
1833 port->mtu = dev_mtu;
1837 /* For non-internal port find new min mtu. */
1838 old_min = p->min_mtu;
1839 port->mtu = dev_mtu;
1840 p->min_mtu = find_min_mtu(p);
1841 if (p->min_mtu == old_min) {
1845 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1846 struct netdev *netdev = ofport->netdev;
1848 if (!strcmp(netdev_get_type(netdev), "internal")) {
1849 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1850 ofport->mtu = p->min_mtu;
1857 ofproto_rule_destroy__(struct rule *rule)
1860 free(rule->actions);
1861 rule->ofproto->ofproto_class->rule_dealloc(rule);
1865 /* This function allows an ofproto implementation to destroy any rules that
1866 * remain when its ->destruct() function is called. The caller must have
1867 * already uninitialized any derived members of 'rule' (step 5 described in the
1868 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1869 * This function implements steps 6 and 7.
1871 * This function should only be called from an ofproto implementation's
1872 * ->destruct() function. It is not suitable elsewhere. */
1874 ofproto_rule_destroy(struct rule *rule)
1876 assert(!rule->pending);
1877 oftable_remove_rule(rule);
1878 ofproto_rule_destroy__(rule);
1881 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1882 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1885 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1887 const union ofp_action *oa;
1890 if (out_port == OFPP_NONE) {
1893 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1894 if (action_outputs_to_port(oa, htons(out_port))) {
1901 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1902 * statistics appropriately. 'packet' must have at least sizeof(struct
1903 * ofp_packet_in) bytes of headroom.
1905 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1906 * with statistics for 'packet' either way.
1908 * Takes ownership of 'packet'. */
1910 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1914 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1916 flow_extract(packet, 0, 0, in_port, &flow);
1917 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1920 /* Returns true if 'rule' should be hidden from the controller.
1922 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1923 * (e.g. by in-band control) and are intentionally hidden from the
1926 rule_is_hidden(const struct rule *rule)
1928 return rule->cr.priority > UINT16_MAX;
1931 static enum oftable_flags
1932 rule_get_flags(const struct rule *rule)
1934 return rule->ofproto->tables[rule->table_id].flags;
1938 rule_is_modifiable(const struct rule *rule)
1940 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1944 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1946 ofconn_send_reply(ofconn, make_echo_reply(oh));
1951 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1953 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1954 struct ofputil_switch_features features;
1955 struct ofport *port;
1959 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1961 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1963 features.datapath_id = ofproto->datapath_id;
1964 features.n_buffers = pktbuf_capacity();
1965 features.n_tables = ofproto->n_tables;
1966 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1967 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1969 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
1972 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
1974 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1975 ofputil_put_switch_features_port(&port->pp, b);
1978 ofconn_send_reply(ofconn, b);
1983 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1985 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1986 struct ofp_switch_config *osc;
1987 enum ofp_config_flags flags;
1991 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1992 flags = ofproto->frag_handling;
1993 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1994 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1996 osc->flags = htons(flags);
1997 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1998 ofconn_send_reply(ofconn, buf);
2004 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2006 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2007 uint16_t flags = ntohs(osc->flags);
2009 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2010 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2011 enum ofp_config_flags cur = ofproto->frag_handling;
2012 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2014 assert((cur & OFPC_FRAG_MASK) == cur);
2016 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2017 ofproto->frag_handling = next;
2019 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2021 ofputil_frag_handling_to_string(next));
2025 ofconn_set_invalid_ttl_to_controller(ofconn,
2026 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2028 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2033 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2034 * error message code for the caller to propagate upward. Otherwise, returns
2037 * The log message mentions 'msg_type'. */
2039 reject_slave_controller(struct ofconn *ofconn)
2041 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2042 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2043 return OFPERR_OFPBRC_EPERM;
2050 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
2052 struct ofproto *p = ofconn_get_ofproto(ofconn);
2053 struct ofputil_packet_out po;
2054 struct ofpbuf *payload;
2058 COVERAGE_INC(ofproto_packet_out);
2060 error = reject_slave_controller(ofconn);
2065 /* Decode message. */
2066 error = ofputil_decode_packet_out(&po, opo);
2072 if (po.buffer_id != UINT32_MAX) {
2073 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2074 if (error || !payload) {
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.actions, po.n_actions);
2086 ofpbuf_delete(payload);
2092 update_port_config(struct ofport *port,
2093 enum ofputil_port_config config,
2094 enum ofputil_port_config mask)
2096 enum ofputil_port_config old_config = port->pp.config;
2097 enum ofputil_port_config toggle;
2099 toggle = (config ^ port->pp.config) & mask;
2100 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2101 if (config & OFPUTIL_PC_PORT_DOWN) {
2102 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2104 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2106 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2109 port->pp.config ^= toggle;
2110 if (port->pp.config != old_config) {
2111 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2116 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2118 struct ofproto *p = ofconn_get_ofproto(ofconn);
2119 struct ofputil_port_mod pm;
2120 struct ofport *port;
2123 error = reject_slave_controller(ofconn);
2128 error = ofputil_decode_port_mod(oh, &pm);
2133 port = ofproto_get_port(p, pm.port_no);
2135 return OFPERR_OFPPMFC_BAD_PORT;
2136 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2137 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2139 update_port_config(port, pm.config, pm.mask);
2141 netdev_set_advertisements(port->netdev, pm.advertise);
2148 handle_desc_stats_request(struct ofconn *ofconn,
2149 const struct ofp_stats_msg *request)
2151 struct ofproto *p = ofconn_get_ofproto(ofconn);
2152 struct ofp_desc_stats *ods;
2155 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2156 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2157 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2158 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2159 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2160 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2161 ofconn_send_reply(ofconn, msg);
2167 handle_table_stats_request(struct ofconn *ofconn,
2168 const struct ofp_stats_msg *request)
2170 struct ofproto *p = ofconn_get_ofproto(ofconn);
2171 struct ofp_table_stats *ots;
2175 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2177 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2178 for (i = 0; i < p->n_tables; i++) {
2179 ots[i].table_id = i;
2180 sprintf(ots[i].name, "table%zu", i);
2181 ots[i].wildcards = htonl(OFPFW10_ALL);
2182 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2183 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2186 p->ofproto_class->get_tables(p, ots);
2188 for (i = 0; i < p->n_tables; i++) {
2189 const struct oftable *table = &p->tables[i];
2192 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2195 if (table->max_flows < ntohl(ots[i].max_entries)) {
2196 ots[i].max_entries = htonl(table->max_flows);
2200 ofconn_send_reply(ofconn, msg);
2205 append_port_stat(struct ofport *port, struct list *replies)
2207 struct netdev_stats stats;
2208 struct ofp_port_stats *ops;
2210 /* Intentionally ignore return value, since errors will set
2211 * 'stats' to all-1s, which is correct for OpenFlow, and
2212 * netdev_get_stats() will log errors. */
2213 ofproto_port_get_stats(port, &stats);
2215 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2216 ops->port_no = htons(port->pp.port_no);
2217 memset(ops->pad, 0, sizeof ops->pad);
2218 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2219 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2220 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2221 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2222 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2223 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2224 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2225 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2226 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2227 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2228 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2229 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2233 handle_port_stats_request(struct ofconn *ofconn,
2234 const struct ofp_port_stats_request *psr)
2236 struct ofproto *p = ofconn_get_ofproto(ofconn);
2237 struct ofport *port;
2238 struct list replies;
2240 ofputil_start_stats_reply(&psr->osm, &replies);
2241 if (psr->port_no != htons(OFPP_NONE)) {
2242 port = ofproto_get_port(p, ntohs(psr->port_no));
2244 append_port_stat(port, &replies);
2247 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2248 append_port_stat(port, &replies);
2252 ofconn_send_replies(ofconn, &replies);
2257 handle_port_desc_stats_request(struct ofconn *ofconn,
2258 const struct ofp_stats_msg *osm)
2260 struct ofproto *p = ofconn_get_ofproto(ofconn);
2261 struct ofport *port;
2262 struct list replies;
2264 ofputil_start_stats_reply(osm, &replies);
2266 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2267 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2268 &port->pp, &replies);
2271 ofconn_send_replies(ofconn, &replies);
2276 calc_flow_duration__(long long int start, long long int now,
2277 uint32_t *sec, uint32_t *nsec)
2279 long long int msecs = now - start;
2280 *sec = msecs / 1000;
2281 *nsec = (msecs % 1000) * (1000 * 1000);
2284 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2285 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2287 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2289 return (table_id == 0xff || table_id < ofproto->n_tables
2291 : OFPERR_NXBRC_BAD_TABLE_ID);
2295 static struct oftable *
2296 next_visible_table(struct ofproto *ofproto, uint8_t table_id)
2298 struct oftable *table;
2300 for (table = &ofproto->tables[table_id];
2301 table < &ofproto->tables[ofproto->n_tables];
2303 if (!(table->flags & OFTABLE_HIDDEN)) {
2311 static struct oftable *
2312 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2314 if (table_id == 0xff) {
2315 return next_visible_table(ofproto, 0);
2316 } else if (table_id < ofproto->n_tables) {
2317 return &ofproto->tables[table_id];
2323 static struct oftable *
2324 next_matching_table(struct ofproto *ofproto,
2325 struct oftable *table, uint8_t table_id)
2327 return (table_id == 0xff
2328 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2332 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2334 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2335 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2337 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2338 * only once, for that table. (This can be used to access tables marked
2341 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2342 * entered at all. (Perhaps you should have validated TABLE_ID with
2343 * check_table_id().)
2345 * All parameters are evaluated multiple times.
2347 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2348 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2350 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2352 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2353 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2354 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2357 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2358 * to 'out_port' are included.
2360 * Hidden rules are always omitted.
2362 * Returns 0 on success, otherwise an OpenFlow error code. */
2364 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2365 const struct cls_rule *match,
2366 ovs_be64 cookie, ovs_be64 cookie_mask,
2367 uint16_t out_port, struct list *rules)
2369 struct oftable *table;
2372 error = check_table_id(ofproto, table_id);
2378 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2379 struct cls_cursor cursor;
2382 cls_cursor_init(&cursor, &table->cls, match);
2383 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2384 if (rule->pending) {
2385 return OFPROTO_POSTPONE;
2387 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2388 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2389 list_push_back(rules, &rule->ofproto_node);
2396 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2397 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2398 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2401 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2402 * to 'out_port' are included.
2404 * Hidden rules are always omitted.
2406 * Returns 0 on success, otherwise an OpenFlow error code. */
2408 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2409 const struct cls_rule *match,
2410 ovs_be64 cookie, ovs_be64 cookie_mask,
2411 uint16_t out_port, struct list *rules)
2413 struct oftable *table;
2416 error = check_table_id(ofproto, table_id);
2422 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2425 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2428 if (rule->pending) {
2429 return OFPROTO_POSTPONE;
2431 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2432 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2433 list_push_back(rules, &rule->ofproto_node);
2440 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2441 * forced into the range of a uint16_t. */
2443 age_secs(long long int age_ms)
2445 return (age_ms < 0 ? 0
2446 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2447 : (unsigned int) age_ms / 1000);
2451 handle_flow_stats_request(struct ofconn *ofconn,
2452 const struct ofp_stats_msg *osm)
2454 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2455 struct ofputil_flow_stats_request fsr;
2456 struct list replies;
2461 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2466 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2467 fsr.cookie, fsr.cookie_mask,
2468 fsr.out_port, &rules);
2473 ofputil_start_stats_reply(osm, &replies);
2474 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2475 long long int now = time_msec();
2476 struct ofputil_flow_stats fs;
2479 fs.cookie = rule->flow_cookie;
2480 fs.table_id = rule->table_id;
2481 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2483 fs.idle_timeout = rule->idle_timeout;
2484 fs.hard_timeout = rule->hard_timeout;
2485 fs.idle_age = age_secs(now - rule->used);
2486 fs.hard_age = age_secs(now - rule->modified);
2487 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2489 fs.actions = rule->actions;
2490 fs.n_actions = rule->n_actions;
2491 ofputil_append_flow_stats_reply(&fs, &replies);
2493 ofconn_send_replies(ofconn, &replies);
2499 flow_stats_ds(struct rule *rule, struct ds *results)
2501 uint64_t packet_count, byte_count;
2503 rule->ofproto->ofproto_class->rule_get_stats(rule,
2504 &packet_count, &byte_count);
2506 if (rule->table_id != 0) {
2507 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2509 ds_put_format(results, "duration=%llds, ",
2510 (time_msec() - rule->created) / 1000);
2511 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2512 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2513 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2514 cls_rule_format(&rule->cr, results);
2515 ds_put_char(results, ',');
2516 if (rule->n_actions > 0) {
2517 ofp_print_actions(results, rule->actions, rule->n_actions);
2519 ds_put_cstr(results, "drop");
2521 ds_put_cstr(results, "\n");
2524 /* Adds a pretty-printed description of all flows to 'results', including
2525 * hidden flows (e.g., set up by in-band control). */
2527 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2529 struct oftable *table;
2531 OFPROTO_FOR_EACH_TABLE (table, p) {
2532 struct cls_cursor cursor;
2535 cls_cursor_init(&cursor, &table->cls, NULL);
2536 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2537 flow_stats_ds(rule, results);
2542 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2543 * '*engine_type' and '*engine_id', respectively. */
2545 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2546 uint8_t *engine_type, uint8_t *engine_id)
2548 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2551 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2552 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2553 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2554 * and -1 if CFM is not enabled on 'port'. */
2556 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2558 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2559 return (ofport && ofproto->ofproto_class->get_cfm_fault
2560 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2564 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2565 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2566 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2567 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2569 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2570 uint16_t ofp_port, const uint64_t **rmps,
2573 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2577 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2578 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2583 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2584 * integer value between 0 and 100 to indicate the health of the port as a
2585 * percentage which is the average of cfm health of all the remote_mpids or
2586 * returns -1 if CFM is not enabled on 'ofport'. */
2588 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2590 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2591 return (ofport && ofproto->ofproto_class->get_cfm_health
2592 ? ofproto->ofproto_class->get_cfm_health(ofport)
2597 handle_aggregate_stats_request(struct ofconn *ofconn,
2598 const struct ofp_stats_msg *osm)
2600 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2601 struct ofputil_flow_stats_request request;
2602 struct ofputil_aggregate_stats stats;
2603 bool unknown_packets, unknown_bytes;
2604 struct ofpbuf *reply;
2609 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2614 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2615 request.cookie, request.cookie_mask,
2616 request.out_port, &rules);
2621 memset(&stats, 0, sizeof stats);
2622 unknown_packets = unknown_bytes = false;
2623 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2624 uint64_t packet_count;
2625 uint64_t byte_count;
2627 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2630 if (packet_count == UINT64_MAX) {
2631 unknown_packets = true;
2633 stats.packet_count += packet_count;
2636 if (byte_count == UINT64_MAX) {
2637 unknown_bytes = true;
2639 stats.byte_count += byte_count;
2644 if (unknown_packets) {
2645 stats.packet_count = UINT64_MAX;
2647 if (unknown_bytes) {
2648 stats.byte_count = UINT64_MAX;
2651 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2652 ofconn_send_reply(ofconn, reply);
2657 struct queue_stats_cbdata {
2658 struct ofport *ofport;
2659 struct list replies;
2663 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2664 const struct netdev_queue_stats *stats)
2666 struct ofp_queue_stats *reply;
2668 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2669 reply->port_no = htons(cbdata->ofport->pp.port_no);
2670 memset(reply->pad, 0, sizeof reply->pad);
2671 reply->queue_id = htonl(queue_id);
2672 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2673 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2674 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2678 handle_queue_stats_dump_cb(uint32_t queue_id,
2679 struct netdev_queue_stats *stats,
2682 struct queue_stats_cbdata *cbdata = cbdata_;
2684 put_queue_stats(cbdata, queue_id, stats);
2688 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2689 struct queue_stats_cbdata *cbdata)
2691 cbdata->ofport = port;
2692 if (queue_id == OFPQ_ALL) {
2693 netdev_dump_queue_stats(port->netdev,
2694 handle_queue_stats_dump_cb, cbdata);
2696 struct netdev_queue_stats stats;
2698 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2699 put_queue_stats(cbdata, queue_id, &stats);
2701 return OFPERR_OFPQOFC_BAD_QUEUE;
2708 handle_queue_stats_request(struct ofconn *ofconn,
2709 const struct ofp_queue_stats_request *qsr)
2711 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2712 struct queue_stats_cbdata cbdata;
2713 unsigned int port_no;
2714 struct ofport *port;
2718 COVERAGE_INC(ofproto_queue_req);
2720 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2722 port_no = ntohs(qsr->port_no);
2723 queue_id = ntohl(qsr->queue_id);
2724 if (port_no == OFPP_ALL) {
2725 error = OFPERR_OFPQOFC_BAD_QUEUE;
2726 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2727 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2732 port = ofproto_get_port(ofproto, port_no);
2734 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2735 : OFPERR_OFPQOFC_BAD_PORT);
2738 ofconn_send_replies(ofconn, &cbdata.replies);
2740 ofpbuf_list_delete(&cbdata.replies);
2747 is_flow_deletion_pending(const struct ofproto *ofproto,
2748 const struct cls_rule *cls_rule,
2751 if (!hmap_is_empty(&ofproto->deletions)) {
2752 struct ofoperation *op;
2754 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2755 cls_rule_hash(cls_rule, table_id),
2756 &ofproto->deletions) {
2757 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2766 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2767 * in which no matching flow already exists in the flow table.
2769 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2770 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2771 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2772 * initiated now but may be retried later.
2774 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2777 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2778 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2780 struct oftable *table;
2781 struct ofopgroup *group;
2782 struct rule *victim;
2786 error = check_table_id(ofproto, fm->table_id);
2792 if (fm->table_id == 0xff) {
2794 if (ofproto->ofproto_class->rule_choose_table) {
2795 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2800 assert(table_id < ofproto->n_tables);
2801 table = &ofproto->tables[table_id];
2803 table = &ofproto->tables[0];
2805 } else if (fm->table_id < ofproto->n_tables) {
2806 table = &ofproto->tables[fm->table_id];
2808 return OFPERR_NXFMFC_BAD_TABLE_ID;
2811 if (table->flags & OFTABLE_READONLY) {
2812 return OFPERR_OFPBRC_EPERM;
2815 /* Check for overlap, if requested. */
2816 if (fm->flags & OFPFF_CHECK_OVERLAP
2817 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2818 return OFPERR_OFPFMFC_OVERLAP;
2821 /* Serialize against pending deletion. */
2822 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2823 return OFPROTO_POSTPONE;
2826 /* Allocate new rule. */
2827 rule = ofproto->ofproto_class->rule_alloc();
2829 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2830 ofproto->name, strerror(error));
2833 rule->ofproto = ofproto;
2835 rule->pending = NULL;
2836 rule->flow_cookie = fm->new_cookie;
2837 rule->created = rule->modified = rule->used = time_msec();
2838 rule->idle_timeout = fm->idle_timeout;
2839 rule->hard_timeout = fm->hard_timeout;
2840 rule->table_id = table - ofproto->tables;
2841 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2842 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2843 rule->n_actions = fm->n_actions;
2844 rule->evictable = true;
2845 rule->eviction_group = NULL;
2847 /* Insert new rule. */
2848 victim = oftable_replace_rule(rule);
2849 if (victim && !rule_is_modifiable(victim)) {
2850 error = OFPERR_OFPBRC_EPERM;
2851 } else if (victim && victim->pending) {
2852 error = OFPROTO_POSTPONE;
2856 if (classifier_count(&table->cls) > table->max_flows) {
2859 was_evictable = rule->evictable;
2860 rule->evictable = false;
2861 evict = choose_rule_to_evict(table);
2862 rule->evictable = was_evictable;
2865 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2867 } else if (evict->pending) {
2868 error = OFPROTO_POSTPONE;
2875 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2876 ofoperation_create(group, rule, OFOPERATION_ADD);
2877 rule->pending->victim = victim;
2879 error = ofproto->ofproto_class->rule_construct(rule);
2881 ofoperation_destroy(rule->pending);
2883 delete_flow__(evict, group);
2885 ofopgroup_submit(group);
2889 /* Back out if an error occurred. */
2891 oftable_substitute_rule(rule, victim);
2892 ofproto_rule_destroy__(rule);
2897 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2899 /* Modifies the rules listed in 'rules', changing their actions to match those
2902 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2905 * Returns 0 on success, otherwise an OpenFlow error code. */
2907 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2908 const struct ofputil_flow_mod *fm,
2909 const struct ofp_header *request, struct list *rules)
2911 struct ofopgroup *group;
2915 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2916 error = OFPERR_OFPBRC_EPERM;
2917 LIST_FOR_EACH (rule, ofproto_node, rules) {
2918 if (rule_is_modifiable(rule)) {
2919 /* At least one rule is modifiable, don't report EPERM error. */
2925 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2926 rule->actions, rule->n_actions)) {
2927 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2928 rule->pending->actions = rule->actions;
2929 rule->pending->n_actions = rule->n_actions;
2930 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2931 rule->n_actions = fm->n_actions;
2932 ofproto->ofproto_class->rule_modify_actions(rule);
2934 rule->modified = time_msec();
2936 if (fm->new_cookie != htonll(UINT64_MAX)) {
2937 rule->flow_cookie = fm->new_cookie;
2940 ofopgroup_submit(group);
2945 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2948 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2951 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2952 const struct ofputil_flow_mod *fm,
2953 const struct ofp_header *request)
2958 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2959 fm->cookie, fm->cookie_mask,
2963 } else if (list_is_empty(&rules)) {
2964 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2966 return modify_flows__(ofproto, ofconn, fm, request, &rules);
2970 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2973 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2976 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2977 const struct ofputil_flow_mod *fm,
2978 const struct ofp_header *request)
2983 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2984 fm->cookie, fm->cookie_mask,
2989 } else if (list_is_empty(&rules)) {
2990 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2992 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2993 fm, request, &rules)
2998 /* OFPFC_DELETE implementation. */
3001 delete_flow__(struct rule *rule, struct ofopgroup *group)
3003 struct ofproto *ofproto = rule->ofproto;
3005 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3007 ofoperation_create(group, rule, OFOPERATION_DELETE);
3008 oftable_remove_rule(rule);
3009 ofproto->ofproto_class->rule_destruct(rule);
3012 /* Deletes the rules listed in 'rules'.
3014 * Returns 0 on success, otherwise an OpenFlow error code. */
3016 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3017 const struct ofp_header *request, struct list *rules)
3019 struct rule *rule, *next;
3020 struct ofopgroup *group;
3022 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3023 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3024 delete_flow__(rule, group);
3026 ofopgroup_submit(group);
3031 /* Implements OFPFC_DELETE. */
3033 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3034 const struct ofputil_flow_mod *fm,
3035 const struct ofp_header *request)
3040 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3041 fm->cookie, fm->cookie_mask,
3042 fm->out_port, &rules);
3043 return (error ? error
3044 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3049 /* Implements OFPFC_DELETE_STRICT. */
3051 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3052 const struct ofputil_flow_mod *fm,
3053 const struct ofp_header *request)
3058 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3059 fm->cookie, fm->cookie_mask,
3060 fm->out_port, &rules);
3061 return (error ? error
3062 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3068 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3070 struct ofputil_flow_removed fr;
3072 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
3077 fr.cookie = rule->flow_cookie;
3079 calc_flow_duration__(rule->created, time_msec(),
3080 &fr.duration_sec, &fr.duration_nsec);
3081 fr.idle_timeout = rule->idle_timeout;
3082 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3085 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3089 ofproto_rule_update_used(struct rule *rule, long long int used)
3091 if (used > rule->used) {
3092 struct eviction_group *evg = rule->eviction_group;
3096 heap_change(&evg->rules, &rule->evg_node,
3097 rule_eviction_priority(rule));
3102 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3103 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3106 * ofproto implementation ->run() functions should use this function to expire
3107 * OpenFlow flows. */
3109 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3111 struct ofproto *ofproto = rule->ofproto;
3112 struct ofopgroup *group;
3114 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3116 ofproto_rule_send_removed(rule, reason);
3118 group = ofopgroup_create_unattached(ofproto);
3119 ofoperation_create(group, rule, OFOPERATION_DELETE);
3120 oftable_remove_rule(rule);
3121 ofproto->ofproto_class->rule_destruct(rule);
3122 ofopgroup_submit(group);
3126 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3128 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3129 struct ofputil_flow_mod fm;
3133 error = reject_slave_controller(ofconn);
3138 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn));
3143 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3144 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3145 if (fm.flags & OFPFF_EMERG) {
3146 /* There isn't a good fit for an error code, so just state that the
3147 * flow table is full. */
3148 return OFPERR_OFPFMFC_ALL_TABLES_FULL;
3151 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3156 /* Record the operation for logging a summary report. */
3157 switch (fm.command) {
3163 case OFPFC_MODIFY_STRICT:
3164 ofproto->n_modify++;
3168 case OFPFC_DELETE_STRICT:
3169 ofproto->n_delete++;
3174 if (ofproto->next_op_report == LLONG_MAX) {
3175 ofproto->first_op = now;
3176 ofproto->next_op_report = MAX(now + 10 * 1000,
3177 ofproto->op_backoff);
3178 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3180 ofproto->last_op = now;
3186 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3187 const struct ofputil_flow_mod *fm,
3188 const struct ofp_header *oh)
3190 if (ofproto->n_pending >= 50) {
3191 assert(!list_is_empty(&ofproto->pending));
3192 return OFPROTO_POSTPONE;
3195 switch (fm->command) {
3197 return add_flow(ofproto, ofconn, fm, oh);
3200 return modify_flows_loose(ofproto, ofconn, fm, oh);
3202 case OFPFC_MODIFY_STRICT:
3203 return modify_flow_strict(ofproto, ofconn, fm, oh);
3206 return delete_flows_loose(ofproto, ofconn, fm, oh);
3208 case OFPFC_DELETE_STRICT:
3209 return delete_flow_strict(ofproto, ofconn, fm, oh);
3212 if (fm->command > 0xff) {
3213 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
3214 "flow_mod_table_id extension is not enabled");
3216 return OFPERR_OFPFMFC_BAD_COMMAND;
3221 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3223 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3224 struct nx_role_request *reply;
3228 role = ntohl(nrr->role);
3229 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3230 && role != NX_ROLE_SLAVE) {
3231 return OFPERR_OFPRRFC_BAD_ROLE;
3234 if (ofconn_get_role(ofconn) != role
3235 && ofconn_has_pending_opgroups(ofconn)) {
3236 return OFPROTO_POSTPONE;
3239 ofconn_set_role(ofconn, role);
3241 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3242 reply->role = htonl(role);
3243 ofconn_send_reply(ofconn, buf);
3249 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3250 const struct ofp_header *oh)
3252 const struct nx_flow_mod_table_id *msg
3253 = (const struct nx_flow_mod_table_id *) oh;
3254 enum ofputil_protocol cur, next;
3256 cur = ofconn_get_protocol(ofconn);
3257 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3258 ofconn_set_protocol(ofconn, next);
3264 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3266 const struct nx_set_flow_format *msg
3267 = (const struct nx_set_flow_format *) oh;
3268 enum ofputil_protocol cur, next;
3269 enum ofputil_protocol next_base;
3271 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3273 return OFPERR_OFPBRC_EPERM;
3276 cur = ofconn_get_protocol(ofconn);
3277 next = ofputil_protocol_set_base(cur, next_base);
3278 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3279 /* Avoid sending async messages in surprising protocol. */
3280 return OFPROTO_POSTPONE;
3283 ofconn_set_protocol(ofconn, next);
3288 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3289 const struct ofp_header *oh)
3291 const struct nx_set_packet_in_format *msg;
3294 msg = (const struct nx_set_packet_in_format *) oh;
3295 format = ntohl(msg->format);
3296 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3297 return OFPERR_OFPBRC_EPERM;
3300 if (format != ofconn_get_packet_in_format(ofconn)
3301 && ofconn_has_pending_opgroups(ofconn)) {
3302 /* Avoid sending async message in surprsing packet in format. */
3303 return OFPROTO_POSTPONE;
3306 ofconn_set_packet_in_format(ofconn, format);
3311 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3313 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3314 uint32_t master[OAM_N_TYPES];
3315 uint32_t slave[OAM_N_TYPES];
3317 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3318 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3319 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3321 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3322 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3323 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3325 ofconn_set_async_config(ofconn, master, slave);
3326 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3327 !ofconn_get_miss_send_len(ofconn)) {
3328 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3335 handle_nxt_set_controller_id(struct ofconn *ofconn,
3336 const struct ofp_header *oh)
3338 const struct nx_controller_id *nci;
3340 nci = (const struct nx_controller_id *) oh;
3341 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3342 return OFPERR_NXBRC_MUST_BE_ZERO;
3345 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3350 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3354 if (ofconn_has_pending_opgroups(ofconn)) {
3355 return OFPROTO_POSTPONE;
3358 make_openflow_xid(sizeof *oh, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3359 ofconn_send_reply(ofconn, buf);
3364 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3366 const struct ofp_header *oh = msg->data;
3367 const struct ofputil_msg_type *type;
3370 error = ofputil_decode_msg_type(oh, &type);
3375 switch (ofputil_msg_type_code(type)) {
3376 /* OpenFlow requests. */
3377 case OFPUTIL_OFPT_ECHO_REQUEST:
3378 return handle_echo_request(ofconn, oh);
3380 case OFPUTIL_OFPT_FEATURES_REQUEST:
3381 return handle_features_request(ofconn, oh);
3383 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3384 return handle_get_config_request(ofconn, oh);
3386 case OFPUTIL_OFPT_SET_CONFIG:
3387 return handle_set_config(ofconn, msg->data);
3389 case OFPUTIL_OFPT_PACKET_OUT:
3390 return handle_packet_out(ofconn, msg->data);
3392 case OFPUTIL_OFPT_PORT_MOD:
3393 return handle_port_mod(ofconn, oh);
3395 case OFPUTIL_OFPT_FLOW_MOD:
3396 return handle_flow_mod(ofconn, oh);
3398 case OFPUTIL_OFPT_BARRIER_REQUEST:
3399 return handle_barrier_request(ofconn, oh);
3401 /* OpenFlow replies. */
3402 case OFPUTIL_OFPT_ECHO_REPLY:
3405 /* Nicira extension requests. */
3406 case OFPUTIL_NXT_ROLE_REQUEST:
3407 return handle_role_request(ofconn, oh);
3409 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3410 return handle_nxt_flow_mod_table_id(ofconn, oh);
3412 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3413 return handle_nxt_set_flow_format(ofconn, oh);
3415 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3416 return handle_nxt_set_packet_in_format(ofconn, oh);
3418 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3419 return handle_nxt_set_controller_id(ofconn, oh);
3421 case OFPUTIL_NXT_FLOW_MOD:
3422 return handle_flow_mod(ofconn, oh);
3424 case OFPUTIL_NXT_FLOW_AGE:
3425 /* Nothing to do. */
3428 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3429 return handle_nxt_set_async_config(ofconn, oh);
3431 /* Statistics requests. */
3432 case OFPUTIL_OFPST_DESC_REQUEST:
3433 return handle_desc_stats_request(ofconn, msg->data);
3435 case OFPUTIL_OFPST_FLOW_REQUEST:
3436 case OFPUTIL_NXST_FLOW_REQUEST:
3437 return handle_flow_stats_request(ofconn, msg->data);
3439 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3440 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3441 return handle_aggregate_stats_request(ofconn, msg->data);
3443 case OFPUTIL_OFPST_TABLE_REQUEST:
3444 return handle_table_stats_request(ofconn, msg->data);
3446 case OFPUTIL_OFPST_PORT_REQUEST:
3447 return handle_port_stats_request(ofconn, msg->data);
3449 case OFPUTIL_OFPST_QUEUE_REQUEST:
3450 return handle_queue_stats_request(ofconn, msg->data);
3452 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3453 return handle_port_desc_stats_request(ofconn, msg->data);
3455 case OFPUTIL_MSG_INVALID:
3456 case OFPUTIL_OFPT_HELLO:
3457 case OFPUTIL_OFPT_ERROR:
3458 case OFPUTIL_OFPT_FEATURES_REPLY:
3459 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3460 case OFPUTIL_OFPT_PACKET_IN:
3461 case OFPUTIL_OFPT_FLOW_REMOVED:
3462 case OFPUTIL_OFPT_PORT_STATUS:
3463 case OFPUTIL_OFPT_BARRIER_REPLY:
3464 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3465 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3466 case OFPUTIL_OFPST_DESC_REPLY:
3467 case OFPUTIL_OFPST_FLOW_REPLY:
3468 case OFPUTIL_OFPST_QUEUE_REPLY:
3469 case OFPUTIL_OFPST_PORT_REPLY:
3470 case OFPUTIL_OFPST_TABLE_REPLY:
3471 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3472 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3473 case OFPUTIL_NXT_ROLE_REPLY:
3474 case OFPUTIL_NXT_FLOW_REMOVED:
3475 case OFPUTIL_NXT_PACKET_IN:
3476 case OFPUTIL_NXST_FLOW_REPLY:
3477 case OFPUTIL_NXST_AGGREGATE_REPLY:
3479 return (oh->type == OFPT10_STATS_REQUEST ||
3480 oh->type == OFPT10_STATS_REPLY
3481 ? OFPERR_OFPBRC_BAD_STAT
3482 : OFPERR_OFPBRC_BAD_TYPE);
3487 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3489 int error = handle_openflow__(ofconn, ofp_msg);
3490 if (error && error != OFPROTO_POSTPONE) {
3491 ofconn_send_error(ofconn, ofp_msg->data, error);
3493 COVERAGE_INC(ofproto_recv_openflow);
3494 return error != OFPROTO_POSTPONE;
3497 /* Asynchronous operations. */
3499 /* Creates and returns a new ofopgroup that is not associated with any
3500 * OpenFlow connection.
3502 * The caller should add operations to the returned group with
3503 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3504 static struct ofopgroup *
3505 ofopgroup_create_unattached(struct ofproto *ofproto)
3507 struct ofopgroup *group = xzalloc(sizeof *group);
3508 group->ofproto = ofproto;
3509 list_init(&group->ofproto_node);
3510 list_init(&group->ops);
3511 list_init(&group->ofconn_node);
3515 /* Creates and returns a new ofopgroup for 'ofproto'.
3517 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3518 * connection. The 'request' and 'buffer_id' arguments are ignored.
3520 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3521 * If the ofopgroup eventually fails, then the error reply will include
3522 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3523 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3525 * The caller should add operations to the returned group with
3526 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3527 static struct ofopgroup *
3528 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3529 const struct ofp_header *request, uint32_t buffer_id)
3531 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3533 size_t request_len = ntohs(request->length);
3535 assert(ofconn_get_ofproto(ofconn) == ofproto);
3537 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3538 group->ofconn = ofconn;
3539 group->request = xmemdup(request, MIN(request_len, 64));
3540 group->buffer_id = buffer_id;
3545 /* Submits 'group' for processing.
3547 * If 'group' contains no operations (e.g. none were ever added, or all of the
3548 * ones that were added completed synchronously), then it is destroyed
3549 * immediately. Otherwise it is added to the ofproto's list of pending
3552 ofopgroup_submit(struct ofopgroup *group)
3554 if (list_is_empty(&group->ops)) {
3555 ofopgroup_destroy(group);
3557 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3558 group->ofproto->n_pending++;
3563 ofopgroup_destroy(struct ofopgroup *group)
3565 assert(list_is_empty(&group->ops));
3566 if (!list_is_empty(&group->ofproto_node)) {
3567 assert(group->ofproto->n_pending > 0);
3568 group->ofproto->n_pending--;
3569 list_remove(&group->ofproto_node);
3571 if (!list_is_empty(&group->ofconn_node)) {
3572 list_remove(&group->ofconn_node);
3574 ofconn_send_error(group->ofconn, group->request, group->error);
3576 connmgr_retry(group->ofproto->connmgr);
3578 free(group->request);
3582 /* Initiates a new operation on 'rule', of the specified 'type', within
3583 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3585 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3586 enum ofoperation_type type)
3588 struct ofproto *ofproto = group->ofproto;
3589 struct ofoperation *op;
3591 assert(!rule->pending);
3593 op = rule->pending = xzalloc(sizeof *op);
3595 list_push_back(&group->ops, &op->group_node);
3598 op->flow_cookie = rule->flow_cookie;
3600 if (type == OFOPERATION_DELETE) {
3601 hmap_insert(&ofproto->deletions, &op->hmap_node,
3602 cls_rule_hash(&rule->cr, rule->table_id));
3607 ofoperation_destroy(struct ofoperation *op)
3609 struct ofopgroup *group = op->group;
3612 op->rule->pending = NULL;
3614 if (op->type == OFOPERATION_DELETE) {
3615 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3617 list_remove(&op->group_node);
3621 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3622 ofopgroup_destroy(group);
3626 /* Indicates that 'op' completed with status 'error', which is either 0 to
3627 * indicate success or an OpenFlow error code on failure.
3629 * If 'error' is 0, indicating success, the operation will be committed
3630 * permanently to the flow table. There is one interesting subcase:
3632 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3633 * the flow table (the "victim" rule) by a new one, then the caller must
3634 * have uninitialized any derived state in the victim rule, as in step 5 in
3635 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3636 * performs steps 6 and 7 for the victim rule, most notably by calling its
3637 * ->rule_dealloc() function.
3639 * If 'error' is nonzero, then generally the operation will be rolled back:
3641 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3642 * restores the original rule. The caller must have uninitialized any
3643 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3644 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3645 * and 7 for the new rule, calling its ->rule_dealloc() function.
3647 * - If 'op' is a "modify flow" operation, ofproto restores the original
3650 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3651 * allowed to fail. It must always succeed.
3653 * Please see the large comment in ofproto/ofproto-provider.h titled
3654 * "Asynchronous Operation Support" for more information. */
3656 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3658 struct ofopgroup *group = op->group;
3659 struct rule *rule = op->rule;
3660 struct ofproto *ofproto = rule->ofproto;
3662 assert(rule->pending == op);
3666 && op->type != OFOPERATION_DELETE
3668 && group->buffer_id != UINT32_MAX
3669 && list_is_singleton(&op->group_node)) {
3670 struct ofpbuf *packet;
3673 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3677 error = rule_execute(rule, in_port, packet);
3680 if (!group->error) {
3681 group->error = error;
3685 case OFOPERATION_ADD:
3687 ofproto_rule_destroy__(op->victim);
3688 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3689 == htons(VLAN_VID_MASK)) {
3690 if (ofproto->vlan_bitmap) {
3691 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3693 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3694 bitmap_set1(ofproto->vlan_bitmap, vid);
3695 ofproto->vlans_changed = true;
3698 ofproto->vlans_changed = true;
3702 oftable_substitute_rule(rule, op->victim);
3703 ofproto_rule_destroy__(rule);
3708 case OFOPERATION_DELETE:
3710 ofproto_rule_destroy__(rule);
3714 case OFOPERATION_MODIFY:
3716 rule->modified = time_msec();
3718 free(rule->actions);
3719 rule->actions = op->actions;
3720 rule->n_actions = op->n_actions;
3728 ofoperation_destroy(op);
3732 ofoperation_get_victim(struct ofoperation *op)
3734 assert(op->type == OFOPERATION_ADD);
3739 pick_datapath_id(const struct ofproto *ofproto)
3741 const struct ofport *port;
3743 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3745 uint8_t ea[ETH_ADDR_LEN];
3748 error = netdev_get_etheraddr(port->netdev, ea);
3750 return eth_addr_to_uint64(ea);
3752 VLOG_WARN("could not get MAC address for %s (%s)",
3753 netdev_get_name(port->netdev), strerror(error));
3755 return ofproto->fallback_dpid;
3759 pick_fallback_dpid(void)
3761 uint8_t ea[ETH_ADDR_LEN];
3762 eth_addr_nicira_random(ea);
3763 return eth_addr_to_uint64(ea);
3766 /* Table overflow policy. */
3768 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3769 * is not configured to evict rules or if the table contains no evictable
3770 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3772 static struct rule *
3773 choose_rule_to_evict(struct oftable *table)
3775 struct eviction_group *evg;
3777 if (!table->eviction_fields) {
3781 /* In the common case, the outer and inner loops here will each be entered
3784 * - The inner loop normally "return"s in its first iteration. If the
3785 * eviction group has any evictable rules, then it always returns in
3788 * - The outer loop only iterates more than once if the largest eviction
3789 * group has no evictable rules.
3791 * - The outer loop can exit only if table's 'max_flows' is all filled up
3792 * by unevictable rules'. */
3793 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
3796 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
3797 if (rule->evictable) {
3806 /* Searches 'ofproto' for tables that have more flows than their configured
3807 * maximum and that have flow eviction enabled, and evicts as many flows as
3808 * necessary and currently feasible from them.
3810 * This triggers only when an OpenFlow table has N flows in it and then the
3811 * client configures a maximum number of flows less than N. */
3813 ofproto_evict(struct ofproto *ofproto)
3815 struct ofopgroup *group;
3816 struct oftable *table;
3818 group = ofopgroup_create_unattached(ofproto);
3819 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
3820 while (classifier_count(&table->cls) > table->max_flows
3821 && table->eviction_fields) {
3824 rule = choose_rule_to_evict(table);
3825 if (!rule || rule->pending) {
3829 ofoperation_create(group, rule, OFOPERATION_DELETE);
3830 oftable_remove_rule(rule);
3831 ofproto->ofproto_class->rule_destruct(rule);
3834 ofopgroup_submit(group);
3837 /* Eviction groups. */
3839 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3840 * rules. The priority contains low-order random bits to ensure that eviction
3841 * groups with the same number of rules are prioritized randomly. */
3843 eviction_group_priority(size_t n_rules)
3845 uint16_t size = MIN(UINT16_MAX, n_rules);
3846 return (size << 16) | random_uint16();
3849 /* Updates 'evg', an eviction_group within 'table', following a change that
3850 * adds or removes rules in 'evg'. */
3852 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
3854 heap_change(&table->eviction_groups_by_size, &evg->size_node,
3855 eviction_group_priority(heap_count(&evg->rules)));
3858 /* Destroys 'evg', an eviction_group within 'table':
3860 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3861 * rules themselves, just removes them from the eviction group.)
3863 * - Removes 'evg' from 'table'.
3867 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
3869 while (!heap_is_empty(&evg->rules)) {
3872 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
3873 rule->eviction_group = NULL;
3875 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
3876 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
3877 heap_destroy(&evg->rules);
3881 /* Removes 'rule' from its eviction group, if any. */
3883 eviction_group_remove_rule(struct rule *rule)
3885 if (rule->eviction_group) {
3886 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3887 struct eviction_group *evg = rule->eviction_group;
3889 rule->eviction_group = NULL;
3890 heap_remove(&evg->rules, &rule->evg_node);
3891 if (heap_is_empty(&evg->rules)) {
3892 eviction_group_destroy(table, evg);
3894 eviction_group_resized(table, evg);
3899 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3900 * returns the hash value. */
3902 eviction_group_hash_rule(struct rule *rule)
3904 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3905 const struct mf_subfield *sf;
3908 hash = table->eviction_group_id_basis;
3909 for (sf = table->eviction_fields;
3910 sf < &table->eviction_fields[table->n_eviction_fields];
3913 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
3914 union mf_value value;
3916 mf_get_value(sf->field, &rule->cr.flow, &value);
3918 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
3920 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
3921 unsigned int start = sf->ofs + sf->n_bits;
3922 bitwise_zero(&value, sf->field->n_bytes, start,
3923 sf->field->n_bytes * 8 - start);
3925 hash = hash_bytes(&value, sf->field->n_bytes, hash);
3927 hash = hash_int(hash, 0);
3934 /* Returns an eviction group within 'table' with the given 'id', creating one
3936 static struct eviction_group *
3937 eviction_group_find(struct oftable *table, uint32_t id)
3939 struct eviction_group *evg;
3941 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
3945 evg = xmalloc(sizeof *evg);
3946 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
3947 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
3948 eviction_group_priority(0));
3949 heap_init(&evg->rules);
3954 /* Returns an eviction priority for 'rule'. The return value should be
3955 * interpreted so that higher priorities make a rule more attractive candidates
3958 rule_eviction_priority(struct rule *rule)
3960 long long int hard_expiration;
3961 long long int idle_expiration;
3962 long long int expiration;
3963 uint32_t expiration_offset;
3965 /* Calculate time of expiration. */
3966 hard_expiration = (rule->hard_timeout
3967 ? rule->modified + rule->hard_timeout * 1000
3969 idle_expiration = (rule->idle_timeout
3970 ? rule->used + rule->idle_timeout * 1000
3972 expiration = MIN(hard_expiration, idle_expiration);
3973 if (expiration == LLONG_MAX) {
3977 /* Calculate the time of expiration as a number of (approximate) seconds
3978 * after program startup.
3980 * This should work OK for program runs that last UINT32_MAX seconds or
3981 * less. Therefore, please restart OVS at least once every 136 years. */
3982 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
3984 /* Invert the expiration offset because we're using a max-heap. */
3985 return UINT32_MAX - expiration_offset;
3988 /* Adds 'rule' to an appropriate eviction group for its oftable's
3989 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
3990 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
3993 * The caller must ensure that 'rule' is not already in an eviction group. */
3995 eviction_group_add_rule(struct rule *rule)
3997 struct ofproto *ofproto = rule->ofproto;
3998 struct oftable *table = &ofproto->tables[rule->table_id];
4000 if (table->eviction_fields
4001 && (rule->hard_timeout || rule->idle_timeout)) {
4002 struct eviction_group *evg;
4004 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4006 rule->eviction_group = evg;
4007 heap_insert(&evg->rules, &rule->evg_node,
4008 rule_eviction_priority(rule));
4009 eviction_group_resized(table, evg);
4015 /* Initializes 'table'. */
4017 oftable_init(struct oftable *table)
4019 memset(table, 0, sizeof *table);
4020 classifier_init(&table->cls);
4021 table->max_flows = UINT_MAX;
4024 /* Destroys 'table', including its classifier and eviction groups.
4026 * The caller is responsible for freeing 'table' itself. */
4028 oftable_destroy(struct oftable *table)
4030 assert(classifier_is_empty(&table->cls));
4031 oftable_disable_eviction(table);
4032 classifier_destroy(&table->cls);
4036 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4037 * string, then 'table' will use its default name.
4039 * This only affects the name exposed for a table exposed through the OpenFlow
4040 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4042 oftable_set_name(struct oftable *table, const char *name)
4044 if (name && name[0]) {
4045 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4046 if (!table->name || strncmp(name, table->name, len)) {
4048 table->name = xmemdup0(name, len);
4056 /* oftables support a choice of two policies when adding a rule would cause the
4057 * number of flows in the table to exceed the configured maximum number: either
4058 * they can refuse to add the new flow or they can evict some existing flow.
4059 * This function configures the former policy on 'table'. */
4061 oftable_disable_eviction(struct oftable *table)
4063 if (table->eviction_fields) {
4064 struct eviction_group *evg, *next;
4066 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4067 &table->eviction_groups_by_id) {
4068 eviction_group_destroy(table, evg);
4070 hmap_destroy(&table->eviction_groups_by_id);
4071 heap_destroy(&table->eviction_groups_by_size);
4073 free(table->eviction_fields);
4074 table->eviction_fields = NULL;
4075 table->n_eviction_fields = 0;
4079 /* oftables support a choice of two policies when adding a rule would cause the
4080 * number of flows in the table to exceed the configured maximum number: either
4081 * they can refuse to add the new flow or they can evict some existing flow.
4082 * This function configures the latter policy on 'table', with fairness based
4083 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4084 * 'n_fields' as 0 disables fairness.) */
4086 oftable_enable_eviction(struct oftable *table,
4087 const struct mf_subfield *fields, size_t n_fields)
4089 struct cls_cursor cursor;
4092 if (table->eviction_fields
4093 && n_fields == table->n_eviction_fields
4095 || !memcmp(fields, table->eviction_fields,
4096 n_fields * sizeof *fields))) {
4101 oftable_disable_eviction(table);
4103 table->n_eviction_fields = n_fields;
4104 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4106 table->eviction_group_id_basis = random_uint32();
4107 hmap_init(&table->eviction_groups_by_id);
4108 heap_init(&table->eviction_groups_by_size);
4110 cls_cursor_init(&cursor, &table->cls, NULL);
4111 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4112 eviction_group_add_rule(rule);
4116 /* Removes 'rule' from the oftable that contains it. */
4118 oftable_remove_rule(struct rule *rule)
4120 struct ofproto *ofproto = rule->ofproto;
4121 struct oftable *table = &ofproto->tables[rule->table_id];
4123 classifier_remove(&table->cls, &rule->cr);
4124 eviction_group_remove_rule(rule);
4127 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4128 * oftable that has an identical cls_rule. Returns the rule that was removed,
4129 * if any, and otherwise NULL. */
4130 static struct rule *
4131 oftable_replace_rule(struct rule *rule)
4133 struct ofproto *ofproto = rule->ofproto;
4134 struct oftable *table = &ofproto->tables[rule->table_id];
4135 struct rule *victim;
4137 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4139 eviction_group_remove_rule(victim);
4141 eviction_group_add_rule(rule);
4145 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4147 oftable_substitute_rule(struct rule *old, struct rule *new)
4150 oftable_replace_rule(new);
4152 oftable_remove_rule(old);
4156 /* unixctl commands. */
4159 ofproto_lookup(const char *name)
4161 struct ofproto *ofproto;
4163 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4165 if (!strcmp(ofproto->name, name)) {
4173 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4174 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4176 struct ofproto *ofproto;
4180 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4181 ds_put_format(&results, "%s\n", ofproto->name);
4183 unixctl_command_reply(conn, ds_cstr(&results));
4184 ds_destroy(&results);
4188 ofproto_unixctl_init(void)
4190 static bool registered;
4196 unixctl_command_register("ofproto/list", "", 0, 0,
4197 ofproto_unixctl_list, NULL);
4200 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4202 * This is deprecated. It is only for compatibility with broken device drivers
4203 * in old versions of Linux that do not properly support VLANs when VLAN
4204 * devices are not used. When broken device drivers are no longer in
4205 * widespread use, we will delete these interfaces. */
4207 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4208 * (exactly) by an OpenFlow rule in 'ofproto'. */
4210 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4212 const struct oftable *oftable;
4214 free(ofproto->vlan_bitmap);
4215 ofproto->vlan_bitmap = bitmap_allocate(4096);
4216 ofproto->vlans_changed = false;
4218 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4219 const struct cls_table *table;
4221 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4222 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4223 == htons(VLAN_VID_MASK)) {
4224 const struct cls_rule *rule;
4226 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4227 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4228 bitmap_set1(vlan_bitmap, vid);
4229 bitmap_set1(ofproto->vlan_bitmap, vid);
4236 /* Returns true if new VLANs have come into use by the flow table since the
4237 * last call to ofproto_get_vlan_usage().
4239 * We don't track when old VLANs stop being used. */
4241 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4243 return ofproto->vlans_changed;
4246 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4247 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4248 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4249 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4250 * then the VLAN device is un-enslaved. */
4252 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4253 uint16_t realdev_ofp_port, int vid)
4255 struct ofport *ofport;
4258 assert(vlandev_ofp_port != realdev_ofp_port);
4260 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4262 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4263 ofproto->name, vlandev_ofp_port);
4267 if (!ofproto->ofproto_class->set_realdev) {
4268 if (!vlandev_ofp_port) {
4271 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4275 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4277 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4278 ofproto->name, vlandev_ofp_port,
4279 netdev_get_name(ofport->netdev), strerror(error));