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"
50 #include "unaligned.h"
54 VLOG_DEFINE_THIS_MODULE(ofproto);
56 COVERAGE_DEFINE(ofproto_error);
57 COVERAGE_DEFINE(ofproto_flush);
58 COVERAGE_DEFINE(ofproto_no_packet_in);
59 COVERAGE_DEFINE(ofproto_packet_out);
60 COVERAGE_DEFINE(ofproto_queue_req);
61 COVERAGE_DEFINE(ofproto_recv_openflow);
62 COVERAGE_DEFINE(ofproto_reinit_ports);
63 COVERAGE_DEFINE(ofproto_uninstallable);
64 COVERAGE_DEFINE(ofproto_update_port);
67 S_OPENFLOW, /* Processing OpenFlow commands. */
68 S_EVICT, /* Evicting flows from over-limit tables. */
69 S_FLUSH, /* Deleting all flow table rules. */
72 enum ofoperation_type {
78 /* A single OpenFlow request can execute any number of operations. The
79 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
80 * ofconn to which an error reply should be sent if necessary.
82 * ofproto initiates some operations internally. These operations are still
83 * assigned to groups but will not have an associated ofconn. */
85 struct ofproto *ofproto; /* Owning ofproto. */
86 struct list ofproto_node; /* In ofproto's "pending" list. */
87 struct list ops; /* List of "struct ofoperation"s. */
89 /* Data needed to send OpenFlow reply on failure or to send a buffered
92 * If list_is_empty(ofconn_node) then this ofopgroup never had an
93 * associated ofconn or its ofconn's connection dropped after it initiated
94 * the operation. In the latter case 'ofconn' is a wild pointer that
95 * refers to freed memory, so the 'ofconn' member must be used only if
96 * !list_is_empty(ofconn_node).
98 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
99 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
100 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
101 uint32_t buffer_id; /* Buffer id from original request. */
102 int error; /* 0 if no error yet, otherwise error code. */
105 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
106 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
107 const struct ofp_header *,
109 static void ofopgroup_submit(struct ofopgroup *);
110 static void ofopgroup_destroy(struct ofopgroup *);
112 /* A single flow table operation. */
114 struct ofopgroup *group; /* Owning group. */
115 struct list group_node; /* In ofopgroup's "ops" list. */
116 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
117 struct rule *rule; /* Rule being operated upon. */
118 enum ofoperation_type type; /* Type of operation. */
119 int status; /* -1 if pending, otherwise 0 or error code. */
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->vlan_bitmap = NULL;
389 ofproto->vlans_changed = false;
390 ofproto->min_mtu = INT_MAX;
392 error = ofproto->ofproto_class->construct(ofproto);
394 VLOG_ERR("failed to open datapath %s: %s",
395 datapath_name, strerror(error));
396 ofproto_destroy__(ofproto);
400 assert(ofproto->n_tables);
402 ofproto->datapath_id = pick_datapath_id(ofproto);
403 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
411 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
413 struct oftable *table;
415 assert(!ofproto->n_tables);
416 assert(n_tables >= 1 && n_tables <= 255);
418 ofproto->n_tables = n_tables;
419 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
420 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
426 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
428 uint64_t old_dpid = p->datapath_id;
429 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
430 if (p->datapath_id != old_dpid) {
431 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
433 /* Force all active connections to reconnect, since there is no way to
434 * notify a controller that the datapath ID has changed. */
435 ofproto_reconnect_controllers(p);
440 ofproto_set_controllers(struct ofproto *p,
441 const struct ofproto_controller *controllers,
442 size_t n_controllers)
444 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
448 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
450 connmgr_set_fail_mode(p->connmgr, fail_mode);
453 /* Drops the connections between 'ofproto' and all of its controllers, forcing
454 * them to reconnect. */
456 ofproto_reconnect_controllers(struct ofproto *ofproto)
458 connmgr_reconnect(ofproto->connmgr);
461 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
462 * in-band control should guarantee access, in the same way that in-band
463 * control guarantees access to OpenFlow controllers. */
465 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
466 const struct sockaddr_in *extras, size_t n)
468 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
471 /* Sets the OpenFlow queue used by flows set up by in-band control on
472 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
473 * flows will use the default queue. */
475 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
477 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
480 /* Sets the number of flows at which eviction from the kernel flow table
483 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
485 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
486 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
488 ofproto->flow_eviction_threshold = threshold;
492 /* If forward_bpdu is true, the NORMAL action will forward frames with
493 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
494 * the NORMAL action will drop these frames. */
496 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
498 bool old_val = ofproto->forward_bpdu;
499 ofproto->forward_bpdu = forward_bpdu;
500 if (old_val != ofproto->forward_bpdu) {
501 if (ofproto->ofproto_class->forward_bpdu_changed) {
502 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
507 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
508 * 'idle_time', in seconds. */
510 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
512 if (ofproto->ofproto_class->set_mac_idle_time) {
513 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
518 ofproto_set_desc(struct ofproto *p,
519 const char *mfr_desc, const char *hw_desc,
520 const char *sw_desc, const char *serial_desc,
523 struct ofp_desc_stats *ods;
526 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
527 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
528 sizeof ods->mfr_desc);
531 p->mfr_desc = xstrdup(mfr_desc);
534 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
535 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
536 sizeof ods->hw_desc);
539 p->hw_desc = xstrdup(hw_desc);
542 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
543 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
544 sizeof ods->sw_desc);
547 p->sw_desc = xstrdup(sw_desc);
550 if (strlen(serial_desc) >= sizeof ods->serial_num) {
551 VLOG_WARN("truncating serial_desc, must be less than %zu "
553 sizeof ods->serial_num);
555 free(p->serial_desc);
556 p->serial_desc = xstrdup(serial_desc);
559 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
560 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
561 sizeof ods->dp_desc);
564 p->dp_desc = xstrdup(dp_desc);
569 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
571 return connmgr_set_snoops(ofproto->connmgr, snoops);
575 ofproto_set_netflow(struct ofproto *ofproto,
576 const struct netflow_options *nf_options)
578 if (nf_options && sset_is_empty(&nf_options->collectors)) {
582 if (ofproto->ofproto_class->set_netflow) {
583 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
585 return nf_options ? EOPNOTSUPP : 0;
590 ofproto_set_sflow(struct ofproto *ofproto,
591 const struct ofproto_sflow_options *oso)
593 if (oso && sset_is_empty(&oso->targets)) {
597 if (ofproto->ofproto_class->set_sflow) {
598 return ofproto->ofproto_class->set_sflow(ofproto, oso);
600 return oso ? EOPNOTSUPP : 0;
604 /* Spanning Tree Protocol (STP) configuration. */
606 /* Configures STP on 'ofproto' using the settings defined in 's'. If
607 * 's' is NULL, disables STP.
609 * Returns 0 if successful, otherwise a positive errno value. */
611 ofproto_set_stp(struct ofproto *ofproto,
612 const struct ofproto_stp_settings *s)
614 return (ofproto->ofproto_class->set_stp
615 ? ofproto->ofproto_class->set_stp(ofproto, s)
619 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
620 * 'enabled' member of 's' is false, then the other members are not
623 * Returns 0 if successful, otherwise a positive errno value. */
625 ofproto_get_stp_status(struct ofproto *ofproto,
626 struct ofproto_stp_status *s)
628 return (ofproto->ofproto_class->get_stp_status
629 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
633 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
634 * in 's'. The caller is responsible for assigning STP port numbers
635 * (using the 'port_num' member in the range of 1 through 255, inclusive)
636 * and ensuring there are no duplicates. If the 's' is NULL, then STP
637 * is disabled on the port.
639 * Returns 0 if successful, otherwise a positive errno value.*/
641 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
642 const struct ofproto_port_stp_settings *s)
644 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
646 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
647 ofproto->name, ofp_port);
651 return (ofproto->ofproto_class->set_stp_port
652 ? ofproto->ofproto_class->set_stp_port(ofport, s)
656 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
657 * 's'. If the 'enabled' member in 's' is false, then the other members
658 * are not meaningful.
660 * Returns 0 if successful, otherwise a positive errno value.*/
662 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
663 struct ofproto_port_stp_status *s)
665 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
667 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
668 "port %"PRIu16, ofproto->name, ofp_port);
672 return (ofproto->ofproto_class->get_stp_port_status
673 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
677 /* Queue DSCP configuration. */
679 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
680 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
681 * to implement QoS. Instead, it is used to implement features which require
682 * knowledge of what queues exist on a port, and some basic information about
685 * Returns 0 if successful, otherwise a positive errno value. */
687 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
688 const struct ofproto_port_queue *queues,
691 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
694 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
695 ofproto->name, ofp_port);
699 return (ofproto->ofproto_class->set_queues
700 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
704 /* Connectivity Fault Management configuration. */
706 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
708 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
710 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
711 if (ofport && ofproto->ofproto_class->set_cfm) {
712 ofproto->ofproto_class->set_cfm(ofport, NULL);
716 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
717 * basic configuration from the configuration members in 'cfm', and the remote
718 * maintenance point ID from remote_mpid. Ignores the statistics members of
721 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
723 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
724 const struct cfm_settings *s)
726 struct ofport *ofport;
729 ofport = ofproto_get_port(ofproto, ofp_port);
731 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
732 ofproto->name, ofp_port);
736 /* XXX: For configuration simplicity, we only support one remote_mpid
737 * outside of the CFM module. It's not clear if this is the correct long
738 * term solution or not. */
739 error = (ofproto->ofproto_class->set_cfm
740 ? ofproto->ofproto_class->set_cfm(ofport, s)
743 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
744 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
749 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
750 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
751 * 0 if LACP partner information is not current (generally indicating a
752 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
754 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
756 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
757 return (ofport && ofproto->ofproto_class->port_is_lacp_current
758 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
764 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
765 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
766 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
767 * configuration plus, if there is more than one slave, a bonding
770 * If 'aux' is already registered then this function updates its configuration
771 * to 's'. Otherwise, this function registers a new bundle.
773 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
776 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
777 const struct ofproto_bundle_settings *s)
779 return (ofproto->ofproto_class->bundle_set
780 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
784 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
785 * If no such bundle has been registered, this has no effect. */
787 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
789 return ofproto_bundle_register(ofproto, aux, NULL);
793 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
794 * If 'aux' is already registered then this function updates its configuration
795 * to 's'. Otherwise, this function registers a new mirror. */
797 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
798 const struct ofproto_mirror_settings *s)
800 return (ofproto->ofproto_class->mirror_set
801 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
805 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
806 * If no mirror has been registered, this has no effect. */
808 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
810 return ofproto_mirror_register(ofproto, aux, NULL);
813 /* Retrieves statistics from mirror associated with client data pointer
814 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
815 * 'bytes', respectively. If a particular counters is not supported,
816 * the appropriate argument is set to UINT64_MAX. */
818 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
819 uint64_t *packets, uint64_t *bytes)
821 if (!ofproto->ofproto_class->mirror_get_stats) {
822 *packets = *bytes = UINT64_MAX;
826 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
830 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
831 * which all packets are flooded, instead of using MAC learning. If
832 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
834 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
837 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
839 return (ofproto->ofproto_class->set_flood_vlans
840 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
844 /* Returns true if 'aux' is a registered bundle that is currently in use as the
845 * output for a mirror. */
847 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
849 return (ofproto->ofproto_class->is_mirror_output_bundle
850 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
854 /* Configuration of OpenFlow tables. */
856 /* Returns the number of OpenFlow tables in 'ofproto'. */
858 ofproto_get_n_tables(const struct ofproto *ofproto)
860 return ofproto->n_tables;
863 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
864 * settings from 's'. 'table_id' must be in the range 0 through the number of
865 * OpenFlow tables in 'ofproto' minus 1, inclusive.
867 * For read-only tables, only the name may be configured. */
869 ofproto_configure_table(struct ofproto *ofproto, int table_id,
870 const struct ofproto_table_settings *s)
872 struct oftable *table;
874 assert(table_id >= 0 && table_id < ofproto->n_tables);
875 table = &ofproto->tables[table_id];
877 oftable_set_name(table, s->name);
879 if (table->flags & OFTABLE_READONLY) {
884 oftable_enable_eviction(table, s->groups, s->n_groups);
886 oftable_disable_eviction(table);
889 table->max_flows = s->max_flows;
890 if (classifier_count(&table->cls) > table->max_flows
891 && table->eviction_fields) {
892 /* 'table' contains more flows than allowed. We might not be able to
893 * evict them right away because of the asynchronous nature of flow
894 * table changes. Schedule eviction for later. */
895 switch (ofproto->state) {
897 ofproto->state = S_EVICT;
901 /* We're already deleting flows, nothing more to do. */
908 ofproto_has_snoops(const struct ofproto *ofproto)
910 return connmgr_has_snoops(ofproto->connmgr);
914 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
916 connmgr_get_snoops(ofproto->connmgr, snoops);
920 ofproto_flush__(struct ofproto *ofproto)
922 struct ofopgroup *group;
923 struct oftable *table;
925 if (ofproto->ofproto_class->flush) {
926 ofproto->ofproto_class->flush(ofproto);
929 group = ofopgroup_create_unattached(ofproto);
930 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
931 struct rule *rule, *next_rule;
932 struct cls_cursor cursor;
934 if (table->flags & OFTABLE_HIDDEN) {
938 cls_cursor_init(&cursor, &table->cls, NULL);
939 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
940 if (!rule->pending) {
941 ofoperation_create(group, rule, OFOPERATION_DELETE);
942 oftable_remove_rule(rule);
943 ofproto->ofproto_class->rule_destruct(rule);
947 ofopgroup_submit(group);
951 ofproto_destroy__(struct ofproto *ofproto)
953 struct oftable *table;
955 assert(list_is_empty(&ofproto->pending));
956 assert(!ofproto->n_pending);
958 connmgr_destroy(ofproto->connmgr);
960 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
963 free(ofproto->mfr_desc);
964 free(ofproto->hw_desc);
965 free(ofproto->sw_desc);
966 free(ofproto->serial_desc);
967 free(ofproto->dp_desc);
968 hmap_destroy(&ofproto->ports);
969 shash_destroy(&ofproto->port_by_name);
971 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
972 oftable_destroy(table);
974 free(ofproto->tables);
976 hmap_destroy(&ofproto->deletions);
978 free(ofproto->vlan_bitmap);
980 ofproto->ofproto_class->dealloc(ofproto);
984 ofproto_destroy(struct ofproto *p)
986 struct ofport *ofport, *next_ofport;
993 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
994 ofport_destroy(ofport);
997 p->ofproto_class->destruct(p);
998 ofproto_destroy__(p);
1001 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1002 * kernel datapath, for example, this destroys the datapath in the kernel, and
1003 * with the netdev-based datapath, it tears down the data structures that
1004 * represent the datapath.
1006 * The datapath should not be currently open as an ofproto. */
1008 ofproto_delete(const char *name, const char *type)
1010 const struct ofproto_class *class = ofproto_class_find__(type);
1011 return (!class ? EAFNOSUPPORT
1012 : !class->del ? EACCES
1013 : class->del(type, name));
1017 process_port_change(struct ofproto *ofproto, int error, char *devname)
1019 if (error == ENOBUFS) {
1020 reinit_ports(ofproto);
1021 } else if (!error) {
1022 update_port(ofproto, devname);
1028 ofproto_run(struct ofproto *p)
1030 struct sset changed_netdevs;
1031 const char *changed_netdev;
1032 struct ofport *ofport;
1035 error = p->ofproto_class->run(p);
1036 if (error && error != EAGAIN) {
1037 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1040 if (p->ofproto_class->port_poll) {
1043 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1044 process_port_change(p, error, devname);
1048 /* Update OpenFlow port status for any port whose netdev has changed.
1050 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1051 * destroyed, so it's not safe to update ports directly from the
1052 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1053 * need this two-phase approach. */
1054 sset_init(&changed_netdevs);
1055 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1056 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1057 if (ofport->change_seq != change_seq) {
1058 ofport->change_seq = change_seq;
1059 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1062 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1063 update_port(p, changed_netdev);
1065 sset_destroy(&changed_netdevs);
1069 connmgr_run(p->connmgr, handle_openflow);
1073 connmgr_run(p->connmgr, NULL);
1075 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1076 p->state = S_OPENFLOW;
1081 connmgr_run(p->connmgr, NULL);
1083 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1084 connmgr_flushed(p->connmgr);
1085 p->state = S_OPENFLOW;
1096 /* Performs periodic activity required by 'ofproto' that needs to be done
1097 * with the least possible latency.
1099 * It makes sense to call this function a couple of times per poll loop, to
1100 * provide a significant performance boost on some benchmarks with the
1101 * ofproto-dpif implementation. */
1103 ofproto_run_fast(struct ofproto *p)
1107 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1108 if (error && error != EAGAIN) {
1109 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1110 p->name, strerror(error));
1116 ofproto_wait(struct ofproto *p)
1118 struct ofport *ofport;
1120 p->ofproto_class->wait(p);
1121 if (p->ofproto_class->port_poll_wait) {
1122 p->ofproto_class->port_poll_wait(p);
1125 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1126 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1127 poll_immediate_wake();
1133 connmgr_wait(p->connmgr, true);
1138 connmgr_wait(p->connmgr, false);
1139 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1140 poll_immediate_wake();
1147 ofproto_is_alive(const struct ofproto *p)
1149 return connmgr_has_controllers(p->connmgr);
1153 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1156 connmgr_get_controller_info(ofproto->connmgr, info);
1160 ofproto_free_ofproto_controller_info(struct shash *info)
1162 connmgr_free_controller_info(info);
1165 /* Makes a deep copy of 'old' into 'port'. */
1167 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1169 port->name = xstrdup(old->name);
1170 port->type = xstrdup(old->type);
1171 port->ofp_port = old->ofp_port;
1174 /* Frees memory allocated to members of 'ofproto_port'.
1176 * Do not call this function on an ofproto_port obtained from
1177 * ofproto_port_dump_next(): that function retains ownership of the data in the
1180 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1182 free(ofproto_port->name);
1183 free(ofproto_port->type);
1186 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1188 * This function provides no status indication. An error status for the entire
1189 * dump operation is provided when it is completed by calling
1190 * ofproto_port_dump_done().
1193 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1194 const struct ofproto *ofproto)
1196 dump->ofproto = ofproto;
1197 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1201 /* Attempts to retrieve another port from 'dump', which must have been created
1202 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1203 * 'port' and returns true. On failure, returns false.
1205 * Failure might indicate an actual error or merely that the last port has been
1206 * dumped. An error status for the entire dump operation is provided when it
1207 * is completed by calling ofproto_port_dump_done().
1209 * The ofproto owns the data stored in 'port'. It will remain valid until at
1210 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1211 * ofproto_port_dump_done(). */
1213 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1214 struct ofproto_port *port)
1216 const struct ofproto *ofproto = dump->ofproto;
1222 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1225 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1231 /* Completes port table dump operation 'dump', which must have been created
1232 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1233 * error-free, otherwise a positive errno value describing the problem. */
1235 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1237 const struct ofproto *ofproto = dump->ofproto;
1239 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1242 return dump->error == EOF ? 0 : dump->error;
1245 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1246 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1247 * is non-null). On failure, returns a positive errno value and sets
1248 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1250 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1251 uint16_t *ofp_portp)
1256 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1258 update_port(ofproto, netdev_get_name(netdev));
1261 *ofp_portp = error ? OFPP_NONE : ofp_port;
1266 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1267 * initializes '*port' appropriately; on failure, returns a positive errno
1270 * The caller owns the data in 'ofproto_port' and must free it with
1271 * ofproto_port_destroy() when it is no longer needed. */
1273 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1274 struct ofproto_port *port)
1278 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1280 memset(port, 0, sizeof *port);
1285 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1286 * Returns 0 if successful, otherwise a positive errno. */
1288 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1290 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1291 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1294 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1295 if (!error && ofport) {
1296 /* 'name' is the netdev's name and update_port() is going to close the
1297 * netdev. Just in case update_port() refers to 'name' after it
1298 * destroys 'ofport', make a copy of it around the update_port()
1300 char *devname = xstrdup(name);
1301 update_port(ofproto, devname);
1307 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1308 * performs the 'n_actions' actions in 'actions'. The new flow will not
1311 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1312 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1313 * controllers; otherwise, it will be hidden.
1315 * The caller retains ownership of 'cls_rule' and 'actions'.
1317 * This is a helper function for in-band control and fail-open. */
1319 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1320 const union ofp_action *actions, size_t n_actions)
1322 const struct rule *rule;
1324 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1325 &ofproto->tables[0].cls, cls_rule));
1326 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1327 actions, n_actions)) {
1328 struct ofputil_flow_mod fm;
1330 memset(&fm, 0, sizeof fm);
1332 fm.buffer_id = UINT32_MAX;
1333 fm.actions = (union ofp_action *) actions;
1334 fm.n_actions = n_actions;
1335 add_flow(ofproto, NULL, &fm, NULL);
1339 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1340 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1341 * operation cannot be initiated now but may be retried later.
1343 * This is a helper function for in-band control and fail-open. */
1345 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1347 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1350 /* Searches for a rule with matching criteria exactly equal to 'target' in
1351 * ofproto's table 0 and, if it finds one, deletes it.
1353 * This is a helper function for in-band control and fail-open. */
1355 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1359 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1360 &ofproto->tables[0].cls, target));
1362 /* No such rule -> success. */
1364 } else if (rule->pending) {
1365 /* An operation on the rule is already pending -> failure.
1366 * Caller must retry later if it's important. */
1369 /* Initiate deletion -> success. */
1370 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1371 ofoperation_create(group, rule, OFOPERATION_DELETE);
1372 oftable_remove_rule(rule);
1373 ofproto->ofproto_class->rule_destruct(rule);
1374 ofopgroup_submit(group);
1380 /* Starts the process of deleting all of the flows from all of ofproto's flow
1381 * tables and then reintroducing the flows required by in-band control and
1382 * fail-open. The process will complete in a later call to ofproto_run(). */
1384 ofproto_flush_flows(struct ofproto *ofproto)
1386 COVERAGE_INC(ofproto_flush);
1387 ofproto->state = S_FLUSH;
1391 reinit_ports(struct ofproto *p)
1393 struct ofproto_port_dump dump;
1394 struct sset devnames;
1395 struct ofport *ofport;
1396 struct ofproto_port ofproto_port;
1397 const char *devname;
1399 COVERAGE_INC(ofproto_reinit_ports);
1401 sset_init(&devnames);
1402 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1403 sset_add(&devnames, netdev_get_name(ofport->netdev));
1405 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1406 sset_add(&devnames, ofproto_port.name);
1409 SSET_FOR_EACH (devname, &devnames) {
1410 update_port(p, devname);
1412 sset_destroy(&devnames);
1415 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1416 * netdev cannot be opened. On success, also fills in 'opp'. */
1417 static struct netdev *
1418 ofport_open(const struct ofproto_port *ofproto_port,
1419 struct ofputil_phy_port *pp)
1421 enum netdev_flags flags;
1422 struct netdev *netdev;
1425 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1427 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1428 "cannot be opened (%s)",
1429 ofproto_port->name, ofproto_port->ofp_port,
1430 ofproto_port->name, strerror(error));
1434 pp->port_no = ofproto_port->ofp_port;
1435 netdev_get_etheraddr(netdev, pp->hw_addr);
1436 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1437 netdev_get_flags(netdev, &flags);
1438 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1439 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1440 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1441 &pp->supported, &pp->peer);
1442 pp->curr_speed = netdev_features_to_bps(pp->curr);
1443 pp->max_speed = netdev_features_to_bps(pp->supported);
1448 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1449 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1452 ofport_equal(const struct ofputil_phy_port *a,
1453 const struct ofputil_phy_port *b)
1455 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1456 && a->state == b->state
1457 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1458 && a->curr == b->curr
1459 && a->advertised == b->advertised
1460 && a->supported == b->supported
1461 && a->peer == b->peer
1462 && a->curr_speed == b->curr_speed
1463 && a->max_speed == b->max_speed);
1466 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1467 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1468 * one with the same name or port number). */
1470 ofport_install(struct ofproto *p,
1471 struct netdev *netdev, const struct ofputil_phy_port *pp)
1473 const char *netdev_name = netdev_get_name(netdev);
1474 struct ofport *ofport;
1477 /* Create ofport. */
1478 ofport = p->ofproto_class->port_alloc();
1483 ofport->ofproto = p;
1484 ofport->netdev = netdev;
1485 ofport->change_seq = netdev_change_seq(netdev);
1487 ofport->ofp_port = pp->port_no;
1489 /* Add port to 'p'. */
1490 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1491 shash_add(&p->port_by_name, netdev_name, ofport);
1493 update_mtu(p, ofport);
1495 /* Let the ofproto_class initialize its private data. */
1496 error = p->ofproto_class->port_construct(ofport);
1500 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1504 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1505 p->name, netdev_name, strerror(error));
1507 ofport_destroy__(ofport);
1509 netdev_close(netdev);
1513 /* Removes 'ofport' from 'p' and destroys it. */
1515 ofport_remove(struct ofport *ofport)
1517 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1519 ofport_destroy(ofport);
1522 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1525 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1527 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1529 ofport_remove(port);
1533 /* Updates 'port' with new 'pp' description.
1535 * Does not handle a name or port number change. The caller must implement
1536 * such a change as a delete followed by an add. */
1538 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1540 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1541 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1542 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1543 port->pp.state = pp->state;
1544 port->pp.curr = pp->curr;
1545 port->pp.advertised = pp->advertised;
1546 port->pp.supported = pp->supported;
1547 port->pp.peer = pp->peer;
1548 port->pp.curr_speed = pp->curr_speed;
1549 port->pp.max_speed = pp->max_speed;
1551 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1554 /* Update OpenFlow 'state' in 'port' and notify controller. */
1556 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1558 if (port->pp.state != state) {
1559 port->pp.state = state;
1560 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1566 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1568 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1570 if (port->ofproto->ofproto_class->set_realdev) {
1571 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1573 if (port->ofproto->ofproto_class->set_stp_port) {
1574 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1576 if (port->ofproto->ofproto_class->set_cfm) {
1577 port->ofproto->ofproto_class->set_cfm(port, NULL);
1579 if (port->ofproto->ofproto_class->bundle_remove) {
1580 port->ofproto->ofproto_class->bundle_remove(port);
1586 ofport_destroy__(struct ofport *port)
1588 struct ofproto *ofproto = port->ofproto;
1589 const char *name = netdev_get_name(port->netdev);
1591 hmap_remove(&ofproto->ports, &port->hmap_node);
1592 shash_delete(&ofproto->port_by_name,
1593 shash_find(&ofproto->port_by_name, name));
1595 netdev_close(port->netdev);
1596 ofproto->ofproto_class->port_dealloc(port);
1600 ofport_destroy(struct ofport *port)
1603 port->ofproto->ofproto_class->port_destruct(port);
1604 ofport_destroy__(port);
1609 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1611 struct ofport *port;
1613 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1614 hash_int(ofp_port, 0), &ofproto->ports) {
1615 if (port->ofp_port == ofp_port) {
1623 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1625 struct ofproto *ofproto = port->ofproto;
1628 if (ofproto->ofproto_class->port_get_stats) {
1629 error = ofproto->ofproto_class->port_get_stats(port, stats);
1638 update_port(struct ofproto *ofproto, const char *name)
1640 struct ofproto_port ofproto_port;
1641 struct ofputil_phy_port pp;
1642 struct netdev *netdev;
1643 struct ofport *port;
1645 COVERAGE_INC(ofproto_update_port);
1647 /* Fetch 'name''s location and properties from the datapath. */
1648 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1649 ? ofport_open(&ofproto_port, &pp)
1652 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1653 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1654 struct netdev *old_netdev = port->netdev;
1656 /* 'name' hasn't changed location. Any properties changed? */
1657 if (!ofport_equal(&port->pp, &pp)) {
1658 ofport_modified(port, &pp);
1661 update_mtu(ofproto, port);
1663 /* Install the newly opened netdev in case it has changed.
1664 * Don't close the old netdev yet in case port_modified has to
1665 * remove a retained reference to it.*/
1666 port->netdev = netdev;
1667 port->change_seq = netdev_change_seq(netdev);
1669 if (port->ofproto->ofproto_class->port_modified) {
1670 port->ofproto->ofproto_class->port_modified(port);
1673 netdev_close(old_netdev);
1675 /* If 'port' is nonnull then its name differs from 'name' and thus
1676 * we should delete it. If we think there's a port named 'name'
1677 * then its port number must be wrong now so delete it too. */
1679 ofport_remove(port);
1681 ofport_remove_with_name(ofproto, name);
1682 ofport_install(ofproto, netdev, &pp);
1685 /* Any port named 'name' is gone now. */
1686 ofport_remove_with_name(ofproto, name);
1688 ofproto_port_destroy(&ofproto_port);
1692 init_ports(struct ofproto *p)
1694 struct ofproto_port_dump dump;
1695 struct ofproto_port ofproto_port;
1697 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1698 uint16_t ofp_port = ofproto_port.ofp_port;
1699 if (ofproto_get_port(p, ofp_port)) {
1700 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1702 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1703 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1706 struct ofputil_phy_port pp;
1707 struct netdev *netdev;
1709 netdev = ofport_open(&ofproto_port, &pp);
1711 ofport_install(p, netdev, &pp);
1719 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1720 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1722 find_min_mtu(struct ofproto *p)
1724 struct ofport *ofport;
1727 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1728 struct netdev *netdev = ofport->netdev;
1731 /* Skip any internal ports, since that's what we're trying to
1733 if (!strcmp(netdev_get_type(netdev), "internal")) {
1737 if (netdev_get_mtu(netdev, &dev_mtu)) {
1740 if (!mtu || dev_mtu < mtu) {
1745 return mtu ? mtu: ETH_PAYLOAD_MAX;
1748 /* Update MTU of all datapath devices on 'p' to the minimum of the
1749 * non-datapath ports in event of 'port' added or changed. */
1751 update_mtu(struct ofproto *p, struct ofport *port)
1753 struct ofport *ofport;
1754 struct netdev *netdev = port->netdev;
1755 int dev_mtu, old_min;
1757 if (netdev_get_mtu(netdev, &dev_mtu)) {
1761 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1762 if (dev_mtu > p->min_mtu) {
1763 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1764 dev_mtu = p->min_mtu;
1767 port->mtu = dev_mtu;
1771 /* For non-internal port find new min mtu. */
1772 old_min = p->min_mtu;
1773 port->mtu = dev_mtu;
1774 p->min_mtu = find_min_mtu(p);
1775 if (p->min_mtu == old_min) {
1779 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1780 struct netdev *netdev = ofport->netdev;
1782 if (!strcmp(netdev_get_type(netdev), "internal")) {
1783 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1784 ofport->mtu = p->min_mtu;
1791 ofproto_rule_destroy__(struct rule *rule)
1794 free(rule->actions);
1795 rule->ofproto->ofproto_class->rule_dealloc(rule);
1799 /* This function allows an ofproto implementation to destroy any rules that
1800 * remain when its ->destruct() function is called. The caller must have
1801 * already uninitialized any derived members of 'rule' (step 5 described in the
1802 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1803 * This function implements steps 6 and 7.
1805 * This function should only be called from an ofproto implementation's
1806 * ->destruct() function. It is not suitable elsewhere. */
1808 ofproto_rule_destroy(struct rule *rule)
1810 assert(!rule->pending);
1811 oftable_remove_rule(rule);
1812 ofproto_rule_destroy__(rule);
1815 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1816 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1819 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1821 const union ofp_action *oa;
1824 if (out_port == OFPP_NONE) {
1827 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1828 if (action_outputs_to_port(oa, htons(out_port))) {
1835 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1836 * statistics appropriately. 'packet' must have at least sizeof(struct
1837 * ofp_packet_in) bytes of headroom.
1839 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1840 * with statistics for 'packet' either way.
1842 * Takes ownership of 'packet'. */
1844 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1848 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1850 flow_extract(packet, 0, 0, in_port, &flow);
1851 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1854 /* Returns true if 'rule' should be hidden from the controller.
1856 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1857 * (e.g. by in-band control) and are intentionally hidden from the
1860 rule_is_hidden(const struct rule *rule)
1862 return rule->cr.priority > UINT16_MAX;
1865 static enum oftable_flags
1866 rule_get_flags(const struct rule *rule)
1868 return rule->ofproto->tables[rule->table_id].flags;
1872 rule_is_modifiable(const struct rule *rule)
1874 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1878 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1880 ofconn_send_reply(ofconn, make_echo_reply(oh));
1885 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1887 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1888 struct ofputil_switch_features features;
1889 struct ofport *port;
1893 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1895 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1897 features.datapath_id = ofproto->datapath_id;
1898 features.n_buffers = pktbuf_capacity();
1899 features.n_tables = ofproto->n_tables;
1900 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1901 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1903 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
1906 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
1908 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1909 ofputil_put_switch_features_port(&port->pp, b);
1912 ofconn_send_reply(ofconn, b);
1917 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1919 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1920 struct ofp_switch_config *osc;
1921 enum ofp_config_flags flags;
1925 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1926 flags = ofproto->frag_handling;
1927 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1928 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1930 osc->flags = htons(flags);
1931 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1932 ofconn_send_reply(ofconn, buf);
1938 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1940 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1941 uint16_t flags = ntohs(osc->flags);
1943 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
1944 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1945 enum ofp_config_flags cur = ofproto->frag_handling;
1946 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
1948 assert((cur & OFPC_FRAG_MASK) == cur);
1950 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
1951 ofproto->frag_handling = next;
1953 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
1955 ofputil_frag_handling_to_string(next));
1959 ofconn_set_invalid_ttl_to_controller(ofconn,
1960 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
1962 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1967 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1968 * error message code for the caller to propagate upward. Otherwise, returns
1971 * The log message mentions 'msg_type'. */
1973 reject_slave_controller(struct ofconn *ofconn)
1975 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1976 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1977 return OFPERR_OFPBRC_EPERM;
1984 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
1986 struct ofproto *p = ofconn_get_ofproto(ofconn);
1987 struct ofputil_packet_out po;
1988 struct ofpbuf *payload;
1992 COVERAGE_INC(ofproto_packet_out);
1994 error = reject_slave_controller(ofconn);
1999 /* Decode message. */
2000 error = ofputil_decode_packet_out(&po, opo);
2006 if (po.buffer_id != UINT32_MAX) {
2007 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2008 if (error || !payload) {
2012 payload = xmalloc(sizeof *payload);
2013 ofpbuf_use_const(payload, po.packet, po.packet_len);
2016 /* Send out packet. */
2017 flow_extract(payload, 0, 0, po.in_port, &flow);
2018 error = p->ofproto_class->packet_out(p, payload, &flow,
2019 po.actions, po.n_actions);
2020 ofpbuf_delete(payload);
2026 update_port_config(struct ofport *port,
2027 enum ofputil_port_config config,
2028 enum ofputil_port_config mask)
2030 enum ofputil_port_config old_config = port->pp.config;
2031 enum ofputil_port_config toggle;
2033 toggle = (config ^ port->pp.config) & mask;
2034 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2035 if (config & OFPUTIL_PC_PORT_DOWN) {
2036 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2038 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2040 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2043 port->pp.config ^= toggle;
2044 if (port->pp.config != old_config) {
2045 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2050 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2052 struct ofproto *p = ofconn_get_ofproto(ofconn);
2053 struct ofputil_port_mod pm;
2054 struct ofport *port;
2057 error = reject_slave_controller(ofconn);
2062 error = ofputil_decode_port_mod(oh, &pm);
2067 port = ofproto_get_port(p, pm.port_no);
2069 return OFPERR_OFPPMFC_BAD_PORT;
2070 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2071 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2073 update_port_config(port, pm.config, pm.mask);
2075 netdev_set_advertisements(port->netdev, pm.advertise);
2082 handle_desc_stats_request(struct ofconn *ofconn,
2083 const struct ofp_stats_msg *request)
2085 struct ofproto *p = ofconn_get_ofproto(ofconn);
2086 struct ofp_desc_stats *ods;
2089 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2090 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2091 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2092 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2093 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2094 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2095 ofconn_send_reply(ofconn, msg);
2101 handle_table_stats_request(struct ofconn *ofconn,
2102 const struct ofp_stats_msg *request)
2104 struct ofproto *p = ofconn_get_ofproto(ofconn);
2105 struct ofp_table_stats *ots;
2109 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2111 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2112 for (i = 0; i < p->n_tables; i++) {
2113 ots[i].table_id = i;
2114 sprintf(ots[i].name, "table%zu", i);
2115 ots[i].wildcards = htonl(OFPFW_ALL);
2116 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2117 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2120 p->ofproto_class->get_tables(p, ots);
2122 for (i = 0; i < p->n_tables; i++) {
2123 const struct oftable *table = &p->tables[i];
2126 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2129 if (table->max_flows < ntohl(ots[i].max_entries)) {
2130 ots[i].max_entries = htonl(table->max_flows);
2134 ofconn_send_reply(ofconn, msg);
2139 append_port_stat(struct ofport *port, struct list *replies)
2141 struct netdev_stats stats;
2142 struct ofp_port_stats *ops;
2144 /* Intentionally ignore return value, since errors will set
2145 * 'stats' to all-1s, which is correct for OpenFlow, and
2146 * netdev_get_stats() will log errors. */
2147 ofproto_port_get_stats(port, &stats);
2149 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2150 ops->port_no = htons(port->pp.port_no);
2151 memset(ops->pad, 0, sizeof ops->pad);
2152 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2153 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2154 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2155 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2156 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2157 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2158 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2159 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2160 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2161 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2162 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2163 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2167 handle_port_stats_request(struct ofconn *ofconn,
2168 const struct ofp_port_stats_request *psr)
2170 struct ofproto *p = ofconn_get_ofproto(ofconn);
2171 struct ofport *port;
2172 struct list replies;
2174 ofputil_start_stats_reply(&psr->osm, &replies);
2175 if (psr->port_no != htons(OFPP_NONE)) {
2176 port = ofproto_get_port(p, ntohs(psr->port_no));
2178 append_port_stat(port, &replies);
2181 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2182 append_port_stat(port, &replies);
2186 ofconn_send_replies(ofconn, &replies);
2191 handle_port_desc_stats_request(struct ofconn *ofconn,
2192 const struct ofp_stats_msg *osm)
2194 struct ofproto *p = ofconn_get_ofproto(ofconn);
2195 struct ofport *port;
2196 struct list replies;
2198 ofputil_start_stats_reply(osm, &replies);
2200 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2201 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2202 &port->pp, &replies);
2205 ofconn_send_replies(ofconn, &replies);
2210 calc_flow_duration__(long long int start, long long int now,
2211 uint32_t *sec, uint32_t *nsec)
2213 long long int msecs = now - start;
2214 *sec = msecs / 1000;
2215 *nsec = (msecs % 1000) * (1000 * 1000);
2218 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2219 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2221 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2223 return (table_id == 0xff || table_id < ofproto->n_tables
2225 : OFPERR_NXBRC_BAD_TABLE_ID);
2229 static struct oftable *
2230 next_visible_table(struct ofproto *ofproto, uint8_t table_id)
2232 struct oftable *table;
2234 for (table = &ofproto->tables[table_id];
2235 table < &ofproto->tables[ofproto->n_tables];
2237 if (!(table->flags & OFTABLE_HIDDEN)) {
2245 static struct oftable *
2246 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2248 if (table_id == 0xff) {
2249 return next_visible_table(ofproto, 0);
2250 } else if (table_id < ofproto->n_tables) {
2251 return &ofproto->tables[table_id];
2257 static struct oftable *
2258 next_matching_table(struct ofproto *ofproto,
2259 struct oftable *table, uint8_t table_id)
2261 return (table_id == 0xff
2262 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2266 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2268 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2269 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2271 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2272 * only once, for that table. (This can be used to access tables marked
2275 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2276 * entered at all. (Perhaps you should have validated TABLE_ID with
2277 * check_table_id().)
2279 * All parameters are evaluated multiple times.
2281 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2282 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2284 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2286 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2287 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2288 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2291 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2292 * to 'out_port' are included.
2294 * Hidden rules are always omitted.
2296 * Returns 0 on success, otherwise an OpenFlow error code. */
2298 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2299 const struct cls_rule *match,
2300 ovs_be64 cookie, ovs_be64 cookie_mask,
2301 uint16_t out_port, struct list *rules)
2303 struct oftable *table;
2306 error = check_table_id(ofproto, table_id);
2312 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2313 struct cls_cursor cursor;
2316 cls_cursor_init(&cursor, &table->cls, match);
2317 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2318 if (rule->pending) {
2319 return OFPROTO_POSTPONE;
2321 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2322 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2323 list_push_back(rules, &rule->ofproto_node);
2330 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2331 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2332 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2335 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2336 * to 'out_port' are included.
2338 * Hidden rules are always omitted.
2340 * Returns 0 on success, otherwise an OpenFlow error code. */
2342 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2343 const struct cls_rule *match,
2344 ovs_be64 cookie, ovs_be64 cookie_mask,
2345 uint16_t out_port, struct list *rules)
2347 struct oftable *table;
2350 error = check_table_id(ofproto, table_id);
2356 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2359 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2362 if (rule->pending) {
2363 return OFPROTO_POSTPONE;
2365 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2366 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2367 list_push_back(rules, &rule->ofproto_node);
2374 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2375 * forced into the range of a uint16_t. */
2377 age_secs(long long int age_ms)
2379 return (age_ms < 0 ? 0
2380 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2381 : (unsigned int) age_ms / 1000);
2385 handle_flow_stats_request(struct ofconn *ofconn,
2386 const struct ofp_stats_msg *osm)
2388 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2389 struct ofputil_flow_stats_request fsr;
2390 struct list replies;
2395 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2400 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2401 fsr.cookie, fsr.cookie_mask,
2402 fsr.out_port, &rules);
2407 ofputil_start_stats_reply(osm, &replies);
2408 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2409 long long int now = time_msec();
2410 struct ofputil_flow_stats fs;
2413 fs.cookie = rule->flow_cookie;
2414 fs.table_id = rule->table_id;
2415 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2417 fs.idle_timeout = rule->idle_timeout;
2418 fs.hard_timeout = rule->hard_timeout;
2419 fs.idle_age = age_secs(now - rule->used);
2420 fs.hard_age = age_secs(now - rule->modified);
2421 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2423 fs.actions = rule->actions;
2424 fs.n_actions = rule->n_actions;
2425 ofputil_append_flow_stats_reply(&fs, &replies);
2427 ofconn_send_replies(ofconn, &replies);
2433 flow_stats_ds(struct rule *rule, struct ds *results)
2435 uint64_t packet_count, byte_count;
2437 rule->ofproto->ofproto_class->rule_get_stats(rule,
2438 &packet_count, &byte_count);
2440 if (rule->table_id != 0) {
2441 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2443 ds_put_format(results, "duration=%llds, ",
2444 (time_msec() - rule->created) / 1000);
2445 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2446 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2447 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2448 cls_rule_format(&rule->cr, results);
2449 ds_put_char(results, ',');
2450 if (rule->n_actions > 0) {
2451 ofp_print_actions(results, rule->actions, rule->n_actions);
2453 ds_put_cstr(results, "drop");
2455 ds_put_cstr(results, "\n");
2458 /* Adds a pretty-printed description of all flows to 'results', including
2459 * hidden flows (e.g., set up by in-band control). */
2461 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2463 struct oftable *table;
2465 OFPROTO_FOR_EACH_TABLE (table, p) {
2466 struct cls_cursor cursor;
2469 cls_cursor_init(&cursor, &table->cls, NULL);
2470 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2471 flow_stats_ds(rule, results);
2476 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2477 * '*engine_type' and '*engine_id', respectively. */
2479 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2480 uint8_t *engine_type, uint8_t *engine_id)
2482 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2485 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2486 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2487 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2488 * and -1 if CFM is not enabled on 'port'. */
2490 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2492 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2493 return (ofport && ofproto->ofproto_class->get_cfm_fault
2494 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2498 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2499 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2500 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2501 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2503 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2504 uint16_t ofp_port, const uint64_t **rmps,
2507 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2511 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2512 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2517 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2518 * integer value between 0 and 100 to indicate the health of the port as a
2519 * percentage which is the average of cfm health of all the remote_mpids or
2520 * returns -1 if CFM is not enabled on 'ofport'. */
2522 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2524 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2525 return (ofport && ofproto->ofproto_class->get_cfm_health
2526 ? ofproto->ofproto_class->get_cfm_health(ofport)
2531 handle_aggregate_stats_request(struct ofconn *ofconn,
2532 const struct ofp_stats_msg *osm)
2534 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2535 struct ofputil_flow_stats_request request;
2536 struct ofputil_aggregate_stats stats;
2537 bool unknown_packets, unknown_bytes;
2538 struct ofpbuf *reply;
2543 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2548 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2549 request.cookie, request.cookie_mask,
2550 request.out_port, &rules);
2555 memset(&stats, 0, sizeof stats);
2556 unknown_packets = unknown_bytes = false;
2557 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2558 uint64_t packet_count;
2559 uint64_t byte_count;
2561 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2564 if (packet_count == UINT64_MAX) {
2565 unknown_packets = true;
2567 stats.packet_count += packet_count;
2570 if (byte_count == UINT64_MAX) {
2571 unknown_bytes = true;
2573 stats.byte_count += byte_count;
2578 if (unknown_packets) {
2579 stats.packet_count = UINT64_MAX;
2581 if (unknown_bytes) {
2582 stats.byte_count = UINT64_MAX;
2585 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2586 ofconn_send_reply(ofconn, reply);
2591 struct queue_stats_cbdata {
2592 struct ofport *ofport;
2593 struct list replies;
2597 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2598 const struct netdev_queue_stats *stats)
2600 struct ofp_queue_stats *reply;
2602 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2603 reply->port_no = htons(cbdata->ofport->pp.port_no);
2604 memset(reply->pad, 0, sizeof reply->pad);
2605 reply->queue_id = htonl(queue_id);
2606 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2607 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2608 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2612 handle_queue_stats_dump_cb(uint32_t queue_id,
2613 struct netdev_queue_stats *stats,
2616 struct queue_stats_cbdata *cbdata = cbdata_;
2618 put_queue_stats(cbdata, queue_id, stats);
2622 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2623 struct queue_stats_cbdata *cbdata)
2625 cbdata->ofport = port;
2626 if (queue_id == OFPQ_ALL) {
2627 netdev_dump_queue_stats(port->netdev,
2628 handle_queue_stats_dump_cb, cbdata);
2630 struct netdev_queue_stats stats;
2632 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2633 put_queue_stats(cbdata, queue_id, &stats);
2639 handle_queue_stats_request(struct ofconn *ofconn,
2640 const struct ofp_queue_stats_request *qsr)
2642 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2643 struct queue_stats_cbdata cbdata;
2644 struct ofport *port;
2645 unsigned int port_no;
2648 COVERAGE_INC(ofproto_queue_req);
2650 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2652 port_no = ntohs(qsr->port_no);
2653 queue_id = ntohl(qsr->queue_id);
2654 if (port_no == OFPP_ALL) {
2655 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2656 handle_queue_stats_for_port(port, queue_id, &cbdata);
2658 } else if (port_no < OFPP_MAX) {
2659 port = ofproto_get_port(ofproto, port_no);
2661 handle_queue_stats_for_port(port, queue_id, &cbdata);
2664 ofpbuf_list_delete(&cbdata.replies);
2665 return OFPERR_OFPQOFC_BAD_PORT;
2667 ofconn_send_replies(ofconn, &cbdata.replies);
2673 is_flow_deletion_pending(const struct ofproto *ofproto,
2674 const struct cls_rule *cls_rule,
2677 if (!hmap_is_empty(&ofproto->deletions)) {
2678 struct ofoperation *op;
2680 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2681 cls_rule_hash(cls_rule, table_id),
2682 &ofproto->deletions) {
2683 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2692 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2693 * in which no matching flow already exists in the flow table.
2695 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2696 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2697 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2698 * initiated now but may be retried later.
2700 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2703 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2704 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2706 struct oftable *table;
2707 struct ofopgroup *group;
2708 struct rule *victim;
2712 error = check_table_id(ofproto, fm->table_id);
2718 if (fm->table_id == 0xff) {
2720 if (ofproto->ofproto_class->rule_choose_table) {
2721 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2726 assert(table_id < ofproto->n_tables);
2727 table = &ofproto->tables[table_id];
2729 table = &ofproto->tables[0];
2731 } else if (fm->table_id < ofproto->n_tables) {
2732 table = &ofproto->tables[fm->table_id];
2734 return OFPERR_NXFMFC_BAD_TABLE_ID;
2737 if (table->flags & OFTABLE_READONLY) {
2738 return OFPERR_OFPBRC_EPERM;
2741 /* Check for overlap, if requested. */
2742 if (fm->flags & OFPFF_CHECK_OVERLAP
2743 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2744 return OFPERR_OFPFMFC_OVERLAP;
2747 /* Serialize against pending deletion. */
2748 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2749 return OFPROTO_POSTPONE;
2752 /* Allocate new rule. */
2753 rule = ofproto->ofproto_class->rule_alloc();
2755 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2756 ofproto->name, strerror(error));
2759 rule->ofproto = ofproto;
2761 rule->pending = NULL;
2762 rule->flow_cookie = fm->cookie;
2763 rule->created = rule->modified = rule->used = time_msec();
2764 rule->idle_timeout = fm->idle_timeout;
2765 rule->hard_timeout = fm->hard_timeout;
2766 rule->table_id = table - ofproto->tables;
2767 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2768 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2769 rule->n_actions = fm->n_actions;
2770 rule->evictable = true;
2771 rule->eviction_group = NULL;
2773 /* Insert new rule. */
2774 victim = oftable_replace_rule(rule);
2775 if (victim && !rule_is_modifiable(victim)) {
2776 error = OFPERR_OFPBRC_EPERM;
2777 } else if (victim && victim->pending) {
2778 error = OFPROTO_POSTPONE;
2782 if (classifier_count(&table->cls) > table->max_flows) {
2785 was_evictable = rule->evictable;
2786 rule->evictable = false;
2787 evict = choose_rule_to_evict(table);
2788 rule->evictable = was_evictable;
2791 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2793 } else if (evict->pending) {
2794 error = OFPROTO_POSTPONE;
2801 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2802 ofoperation_create(group, rule, OFOPERATION_ADD);
2803 rule->pending->victim = victim;
2805 error = ofproto->ofproto_class->rule_construct(rule);
2807 ofoperation_destroy(rule->pending);
2809 delete_flow__(evict, group);
2811 ofopgroup_submit(group);
2815 /* Back out if an error occurred. */
2817 oftable_substitute_rule(rule, victim);
2818 ofproto_rule_destroy__(rule);
2823 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2825 /* Modifies the rules listed in 'rules', changing their actions to match those
2828 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2831 * Returns 0 on success, otherwise an OpenFlow error code. */
2833 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2834 const struct ofputil_flow_mod *fm,
2835 const struct ofp_header *request, struct list *rules)
2837 struct ofopgroup *group;
2841 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2842 error = OFPERR_OFPBRC_EPERM;
2843 LIST_FOR_EACH (rule, ofproto_node, rules) {
2844 if (rule_is_modifiable(rule)) {
2845 /* At least one rule is modifiable, don't report EPERM error. */
2851 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2852 rule->actions, rule->n_actions)) {
2853 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2854 rule->pending->actions = rule->actions;
2855 rule->pending->n_actions = rule->n_actions;
2856 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2857 rule->n_actions = fm->n_actions;
2858 ofproto->ofproto_class->rule_modify_actions(rule);
2860 rule->modified = time_msec();
2862 rule->flow_cookie = fm->cookie;
2864 ofopgroup_submit(group);
2869 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2872 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2875 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2876 const struct ofputil_flow_mod *fm,
2877 const struct ofp_header *request)
2882 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2883 fm->cookie, fm->cookie_mask,
2885 return (error ? error
2886 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2887 : modify_flows__(ofproto, ofconn, fm, request, &rules));
2890 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2893 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2896 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2897 const struct ofputil_flow_mod *fm,
2898 const struct ofp_header *request)
2903 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2904 fm->cookie, fm->cookie_mask,
2906 return (error ? error
2907 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2908 : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2909 fm, request, &rules)
2913 /* OFPFC_DELETE implementation. */
2916 delete_flow__(struct rule *rule, struct ofopgroup *group)
2918 struct ofproto *ofproto = rule->ofproto;
2920 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2922 ofoperation_create(group, rule, OFOPERATION_DELETE);
2923 oftable_remove_rule(rule);
2924 ofproto->ofproto_class->rule_destruct(rule);
2927 /* Deletes the rules listed in 'rules'.
2929 * Returns 0 on success, otherwise an OpenFlow error code. */
2931 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2932 const struct ofp_header *request, struct list *rules)
2934 struct rule *rule, *next;
2935 struct ofopgroup *group;
2937 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
2938 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2939 delete_flow__(rule, group);
2941 ofopgroup_submit(group);
2946 /* Implements OFPFC_DELETE. */
2948 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2949 const struct ofputil_flow_mod *fm,
2950 const struct ofp_header *request)
2955 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2956 fm->cookie, fm->cookie_mask,
2957 fm->out_port, &rules);
2958 return (error ? error
2959 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
2964 /* Implements OFPFC_DELETE_STRICT. */
2966 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2967 const struct ofputil_flow_mod *fm,
2968 const struct ofp_header *request)
2973 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2974 fm->cookie, fm->cookie_mask,
2975 fm->out_port, &rules);
2976 return (error ? error
2977 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
2983 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2985 struct ofputil_flow_removed fr;
2987 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2992 fr.cookie = rule->flow_cookie;
2994 calc_flow_duration__(rule->created, time_msec(),
2995 &fr.duration_sec, &fr.duration_nsec);
2996 fr.idle_timeout = rule->idle_timeout;
2997 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3000 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3004 ofproto_rule_update_used(struct rule *rule, long long int used)
3006 if (used > rule->used) {
3007 struct eviction_group *evg = rule->eviction_group;
3011 heap_change(&evg->rules, &rule->evg_node,
3012 rule_eviction_priority(rule));
3017 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3018 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3021 * ofproto implementation ->run() functions should use this function to expire
3022 * OpenFlow flows. */
3024 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3026 struct ofproto *ofproto = rule->ofproto;
3027 struct ofopgroup *group;
3029 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3031 ofproto_rule_send_removed(rule, reason);
3033 group = ofopgroup_create_unattached(ofproto);
3034 ofoperation_create(group, rule, OFOPERATION_DELETE);
3035 oftable_remove_rule(rule);
3036 ofproto->ofproto_class->rule_destruct(rule);
3037 ofopgroup_submit(group);
3041 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3043 struct ofputil_flow_mod fm;
3046 error = reject_slave_controller(ofconn);
3051 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn));
3056 /* We do not support the emergency flow cache. It will hopefully get
3057 * dropped from OpenFlow in the near future. */
3058 if (fm.flags & OFPFF_EMERG) {
3059 /* There isn't a good fit for an error code, so just state that the
3060 * flow table is full. */
3061 return OFPERR_OFPFMFC_ALL_TABLES_FULL;
3064 return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3068 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3069 const struct ofputil_flow_mod *fm,
3070 const struct ofp_header *oh)
3072 if (ofproto->n_pending >= 50) {
3073 assert(!list_is_empty(&ofproto->pending));
3074 return OFPROTO_POSTPONE;
3077 switch (fm->command) {
3079 return add_flow(ofproto, ofconn, fm, oh);
3082 return modify_flows_loose(ofproto, ofconn, fm, oh);
3084 case OFPFC_MODIFY_STRICT:
3085 return modify_flow_strict(ofproto, ofconn, fm, oh);
3088 return delete_flows_loose(ofproto, ofconn, fm, oh);
3090 case OFPFC_DELETE_STRICT:
3091 return delete_flow_strict(ofproto, ofconn, fm, oh);
3094 if (fm->command > 0xff) {
3095 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
3096 "flow_mod_table_id extension is not enabled");
3098 return OFPERR_OFPFMFC_BAD_COMMAND;
3103 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3105 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3106 struct nx_role_request *reply;
3110 role = ntohl(nrr->role);
3111 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3112 && role != NX_ROLE_SLAVE) {
3113 return OFPERR_OFPRRFC_BAD_ROLE;
3116 if (ofconn_get_role(ofconn) != role
3117 && ofconn_has_pending_opgroups(ofconn)) {
3118 return OFPROTO_POSTPONE;
3121 ofconn_set_role(ofconn, role);
3123 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3124 reply->role = htonl(role);
3125 ofconn_send_reply(ofconn, buf);
3131 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3132 const struct ofp_header *oh)
3134 const struct nx_flow_mod_table_id *msg
3135 = (const struct nx_flow_mod_table_id *) oh;
3136 enum ofputil_protocol cur, next;
3138 cur = ofconn_get_protocol(ofconn);
3139 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3140 ofconn_set_protocol(ofconn, next);
3146 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3148 const struct nx_set_flow_format *msg
3149 = (const struct nx_set_flow_format *) oh;
3150 enum ofputil_protocol cur, next;
3151 enum ofputil_protocol next_base;
3153 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3155 return OFPERR_OFPBRC_EPERM;
3158 cur = ofconn_get_protocol(ofconn);
3159 next = ofputil_protocol_set_base(cur, next_base);
3160 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3161 /* Avoid sending async messages in surprising protocol. */
3162 return OFPROTO_POSTPONE;
3165 ofconn_set_protocol(ofconn, next);
3170 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3171 const struct ofp_header *oh)
3173 const struct nx_set_packet_in_format *msg;
3176 msg = (const struct nx_set_packet_in_format *) oh;
3177 format = ntohl(msg->format);
3178 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3179 return OFPERR_OFPBRC_EPERM;
3182 if (format != ofconn_get_packet_in_format(ofconn)
3183 && ofconn_has_pending_opgroups(ofconn)) {
3184 /* Avoid sending async message in surprsing packet in format. */
3185 return OFPROTO_POSTPONE;
3188 ofconn_set_packet_in_format(ofconn, format);
3193 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3195 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3196 uint32_t master[OAM_N_TYPES];
3197 uint32_t slave[OAM_N_TYPES];
3199 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3200 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3201 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3203 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3204 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3205 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3207 ofconn_set_async_config(ofconn, master, slave);
3213 handle_nxt_set_controller_id(struct ofconn *ofconn,
3214 const struct ofp_header *oh)
3216 const struct nx_controller_id *nci;
3218 nci = (const struct nx_controller_id *) oh;
3219 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3220 return OFPERR_NXBRC_MUST_BE_ZERO;
3223 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3228 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3230 struct ofp_header *ob;
3233 if (ofconn_has_pending_opgroups(ofconn)) {
3234 return OFPROTO_POSTPONE;
3237 ob = make_openflow_xid(sizeof *ob, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3238 ofconn_send_reply(ofconn, buf);
3243 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3245 const struct ofp_header *oh = msg->data;
3246 const struct ofputil_msg_type *type;
3249 error = ofputil_decode_msg_type(oh, &type);
3254 switch (ofputil_msg_type_code(type)) {
3255 /* OpenFlow requests. */
3256 case OFPUTIL_OFPT_ECHO_REQUEST:
3257 return handle_echo_request(ofconn, oh);
3259 case OFPUTIL_OFPT_FEATURES_REQUEST:
3260 return handle_features_request(ofconn, oh);
3262 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3263 return handle_get_config_request(ofconn, oh);
3265 case OFPUTIL_OFPT_SET_CONFIG:
3266 return handle_set_config(ofconn, msg->data);
3268 case OFPUTIL_OFPT_PACKET_OUT:
3269 return handle_packet_out(ofconn, msg->data);
3271 case OFPUTIL_OFPT_PORT_MOD:
3272 return handle_port_mod(ofconn, oh);
3274 case OFPUTIL_OFPT_FLOW_MOD:
3275 return handle_flow_mod(ofconn, oh);
3277 case OFPUTIL_OFPT_BARRIER_REQUEST:
3278 return handle_barrier_request(ofconn, oh);
3280 /* OpenFlow replies. */
3281 case OFPUTIL_OFPT_ECHO_REPLY:
3284 /* Nicira extension requests. */
3285 case OFPUTIL_NXT_ROLE_REQUEST:
3286 return handle_role_request(ofconn, oh);
3288 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3289 return handle_nxt_flow_mod_table_id(ofconn, oh);
3291 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3292 return handle_nxt_set_flow_format(ofconn, oh);
3294 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3295 return handle_nxt_set_packet_in_format(ofconn, oh);
3297 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3298 return handle_nxt_set_controller_id(ofconn, oh);
3300 case OFPUTIL_NXT_FLOW_MOD:
3301 return handle_flow_mod(ofconn, oh);
3303 case OFPUTIL_NXT_FLOW_AGE:
3304 /* Nothing to do. */
3307 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3308 return handle_nxt_set_async_config(ofconn, oh);
3310 /* Statistics requests. */
3311 case OFPUTIL_OFPST_DESC_REQUEST:
3312 return handle_desc_stats_request(ofconn, msg->data);
3314 case OFPUTIL_OFPST_FLOW_REQUEST:
3315 case OFPUTIL_NXST_FLOW_REQUEST:
3316 return handle_flow_stats_request(ofconn, msg->data);
3318 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3319 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3320 return handle_aggregate_stats_request(ofconn, msg->data);
3322 case OFPUTIL_OFPST_TABLE_REQUEST:
3323 return handle_table_stats_request(ofconn, msg->data);
3325 case OFPUTIL_OFPST_PORT_REQUEST:
3326 return handle_port_stats_request(ofconn, msg->data);
3328 case OFPUTIL_OFPST_QUEUE_REQUEST:
3329 return handle_queue_stats_request(ofconn, msg->data);
3331 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3332 return handle_port_desc_stats_request(ofconn, msg->data);
3334 case OFPUTIL_MSG_INVALID:
3335 case OFPUTIL_OFPT_HELLO:
3336 case OFPUTIL_OFPT_ERROR:
3337 case OFPUTIL_OFPT_FEATURES_REPLY:
3338 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3339 case OFPUTIL_OFPT_PACKET_IN:
3340 case OFPUTIL_OFPT_FLOW_REMOVED:
3341 case OFPUTIL_OFPT_PORT_STATUS:
3342 case OFPUTIL_OFPT_BARRIER_REPLY:
3343 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3344 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3345 case OFPUTIL_OFPST_DESC_REPLY:
3346 case OFPUTIL_OFPST_FLOW_REPLY:
3347 case OFPUTIL_OFPST_QUEUE_REPLY:
3348 case OFPUTIL_OFPST_PORT_REPLY:
3349 case OFPUTIL_OFPST_TABLE_REPLY:
3350 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3351 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3352 case OFPUTIL_NXT_ROLE_REPLY:
3353 case OFPUTIL_NXT_FLOW_REMOVED:
3354 case OFPUTIL_NXT_PACKET_IN:
3355 case OFPUTIL_NXST_FLOW_REPLY:
3356 case OFPUTIL_NXST_AGGREGATE_REPLY:
3358 return (oh->type == OFPT10_STATS_REQUEST ||
3359 oh->type == OFPT10_STATS_REPLY
3360 ? OFPERR_OFPBRC_BAD_STAT
3361 : OFPERR_OFPBRC_BAD_TYPE);
3366 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3368 int error = handle_openflow__(ofconn, ofp_msg);
3369 if (error && error != OFPROTO_POSTPONE) {
3370 ofconn_send_error(ofconn, ofp_msg->data, error);
3372 COVERAGE_INC(ofproto_recv_openflow);
3373 return error != OFPROTO_POSTPONE;
3376 /* Asynchronous operations. */
3378 /* Creates and returns a new ofopgroup that is not associated with any
3379 * OpenFlow connection.
3381 * The caller should add operations to the returned group with
3382 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3383 static struct ofopgroup *
3384 ofopgroup_create_unattached(struct ofproto *ofproto)
3386 struct ofopgroup *group = xzalloc(sizeof *group);
3387 group->ofproto = ofproto;
3388 list_init(&group->ofproto_node);
3389 list_init(&group->ops);
3390 list_init(&group->ofconn_node);
3394 /* Creates and returns a new ofopgroup for 'ofproto'.
3396 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3397 * connection. The 'request' and 'buffer_id' arguments are ignored.
3399 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3400 * If the ofopgroup eventually fails, then the error reply will include
3401 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3402 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3404 * The caller should add operations to the returned group with
3405 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3406 static struct ofopgroup *
3407 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3408 const struct ofp_header *request, uint32_t buffer_id)
3410 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3412 size_t request_len = ntohs(request->length);
3414 assert(ofconn_get_ofproto(ofconn) == ofproto);
3416 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3417 group->ofconn = ofconn;
3418 group->request = xmemdup(request, MIN(request_len, 64));
3419 group->buffer_id = buffer_id;
3424 /* Submits 'group' for processing.
3426 * If 'group' contains no operations (e.g. none were ever added, or all of the
3427 * ones that were added completed synchronously), then it is destroyed
3428 * immediately. Otherwise it is added to the ofproto's list of pending
3431 ofopgroup_submit(struct ofopgroup *group)
3433 if (list_is_empty(&group->ops)) {
3434 ofopgroup_destroy(group);
3436 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3437 group->ofproto->n_pending++;
3442 ofopgroup_destroy(struct ofopgroup *group)
3444 assert(list_is_empty(&group->ops));
3445 if (!list_is_empty(&group->ofproto_node)) {
3446 assert(group->ofproto->n_pending > 0);
3447 group->ofproto->n_pending--;
3448 list_remove(&group->ofproto_node);
3450 if (!list_is_empty(&group->ofconn_node)) {
3451 list_remove(&group->ofconn_node);
3453 ofconn_send_error(group->ofconn, group->request, group->error);
3455 connmgr_retry(group->ofproto->connmgr);
3457 free(group->request);
3461 /* Initiates a new operation on 'rule', of the specified 'type', within
3462 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3464 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3465 enum ofoperation_type type)
3467 struct ofoperation *op;
3469 assert(!rule->pending);
3471 op = rule->pending = xzalloc(sizeof *op);
3473 list_push_back(&group->ops, &op->group_node);
3477 op->flow_cookie = rule->flow_cookie;
3479 if (type == OFOPERATION_DELETE) {
3480 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
3481 cls_rule_hash(&rule->cr, rule->table_id));
3486 ofoperation_destroy(struct ofoperation *op)
3488 struct ofopgroup *group = op->group;
3491 op->rule->pending = NULL;
3493 if (op->type == OFOPERATION_DELETE) {
3494 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3496 list_remove(&op->group_node);
3500 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3501 ofopgroup_destroy(group);
3505 /* Indicates that 'op' completed with status 'error', which is either 0 to
3506 * indicate success or an OpenFlow error code on failure.
3508 * If 'error' is 0, indicating success, the operation will be committed
3509 * permanently to the flow table. There is one interesting subcase:
3511 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3512 * the flow table (the "victim" rule) by a new one, then the caller must
3513 * have uninitialized any derived state in the victim rule, as in step 5 in
3514 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3515 * performs steps 6 and 7 for the victim rule, most notably by calling its
3516 * ->rule_dealloc() function.
3518 * If 'error' is nonzero, then generally the operation will be rolled back:
3520 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3521 * restores the original rule. The caller must have uninitialized any
3522 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3523 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3524 * and 7 for the new rule, calling its ->rule_dealloc() function.
3526 * - If 'op' is a "modify flow" operation, ofproto restores the original
3529 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3530 * allowed to fail. It must always succeed.
3532 * Please see the large comment in ofproto/ofproto-provider.h titled
3533 * "Asynchronous Operation Support" for more information. */
3535 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3537 struct ofopgroup *group = op->group;
3538 struct rule *rule = op->rule;
3539 struct ofproto *ofproto = rule->ofproto;
3541 assert(rule->pending == op);
3542 assert(op->status < 0);
3546 && op->type != OFOPERATION_DELETE
3548 && group->buffer_id != UINT32_MAX
3549 && list_is_singleton(&op->group_node)) {
3550 struct ofpbuf *packet;
3553 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3557 error = rule_execute(rule, in_port, packet);
3560 if (!group->error) {
3561 group->error = error;
3565 case OFOPERATION_ADD:
3567 ofproto_rule_destroy__(op->victim);
3568 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3569 == htons(VLAN_VID_MASK)) {
3570 if (ofproto->vlan_bitmap) {
3571 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3573 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3574 bitmap_set1(ofproto->vlan_bitmap, vid);
3575 ofproto->vlans_changed = true;
3578 ofproto->vlans_changed = true;
3582 oftable_substitute_rule(rule, op->victim);
3583 ofproto_rule_destroy__(rule);
3587 case OFOPERATION_DELETE:
3589 ofproto_rule_destroy__(rule);
3593 case OFOPERATION_MODIFY:
3595 rule->modified = time_msec();
3597 free(rule->actions);
3598 rule->actions = op->actions;
3599 rule->n_actions = op->n_actions;
3607 ofoperation_destroy(op);
3611 ofoperation_get_victim(struct ofoperation *op)
3613 assert(op->type == OFOPERATION_ADD);
3618 pick_datapath_id(const struct ofproto *ofproto)
3620 const struct ofport *port;
3622 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3624 uint8_t ea[ETH_ADDR_LEN];
3627 error = netdev_get_etheraddr(port->netdev, ea);
3629 return eth_addr_to_uint64(ea);
3631 VLOG_WARN("could not get MAC address for %s (%s)",
3632 netdev_get_name(port->netdev), strerror(error));
3634 return ofproto->fallback_dpid;
3638 pick_fallback_dpid(void)
3640 uint8_t ea[ETH_ADDR_LEN];
3641 eth_addr_nicira_random(ea);
3642 return eth_addr_to_uint64(ea);
3645 /* Table overflow policy. */
3647 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3648 * is not configured to evict rules or if the table contains no evictable
3649 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3651 static struct rule *
3652 choose_rule_to_evict(struct oftable *table)
3654 struct eviction_group *evg;
3656 if (!table->eviction_fields) {
3660 /* In the common case, the outer and inner loops here will each be entered
3663 * - The inner loop normally "return"s in its first iteration. If the
3664 * eviction group has any evictable rules, then it always returns in
3667 * - The outer loop only iterates more than once if the largest eviction
3668 * group has no evictable rules.
3670 * - The outer loop can exit only if table's 'max_flows' is all filled up
3671 * by unevictable rules'. */
3672 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
3675 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
3676 if (rule->evictable) {
3685 /* Searches 'ofproto' for tables that have more flows than their configured
3686 * maximum and that have flow eviction enabled, and evicts as many flows as
3687 * necessary and currently feasible from them.
3689 * This triggers only when an OpenFlow table has N flows in it and then the
3690 * client configures a maximum number of flows less than N. */
3692 ofproto_evict(struct ofproto *ofproto)
3694 struct ofopgroup *group;
3695 struct oftable *table;
3697 group = ofopgroup_create_unattached(ofproto);
3698 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
3699 while (classifier_count(&table->cls) > table->max_flows
3700 && table->eviction_fields) {
3703 rule = choose_rule_to_evict(table);
3704 if (!rule || rule->pending) {
3708 ofoperation_create(group, rule, OFOPERATION_DELETE);
3709 oftable_remove_rule(rule);
3710 ofproto->ofproto_class->rule_destruct(rule);
3713 ofopgroup_submit(group);
3716 /* Eviction groups. */
3718 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3719 * rules. The priority contains low-order random bits to ensure that eviction
3720 * groups with the same number of rules are prioritized randomly. */
3722 eviction_group_priority(size_t n_rules)
3724 uint16_t size = MIN(UINT16_MAX, n_rules);
3725 return (size << 16) | random_uint16();
3728 /* Updates 'evg', an eviction_group within 'table', following a change that
3729 * adds or removes rules in 'evg'. */
3731 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
3733 heap_change(&table->eviction_groups_by_size, &evg->size_node,
3734 eviction_group_priority(heap_count(&evg->rules)));
3737 /* Destroys 'evg', an eviction_group within 'table':
3739 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3740 * rules themselves, just removes them from the eviction group.)
3742 * - Removes 'evg' from 'table'.
3746 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
3748 while (!heap_is_empty(&evg->rules)) {
3751 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
3752 rule->eviction_group = NULL;
3754 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
3755 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
3756 heap_destroy(&evg->rules);
3760 /* Removes 'rule' from its eviction group, if any. */
3762 eviction_group_remove_rule(struct rule *rule)
3764 if (rule->eviction_group) {
3765 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3766 struct eviction_group *evg = rule->eviction_group;
3768 rule->eviction_group = NULL;
3769 heap_remove(&evg->rules, &rule->evg_node);
3770 if (heap_is_empty(&evg->rules)) {
3771 eviction_group_destroy(table, evg);
3773 eviction_group_resized(table, evg);
3778 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3779 * returns the hash value. */
3781 eviction_group_hash_rule(struct rule *rule)
3783 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3784 const struct mf_subfield *sf;
3787 hash = table->eviction_group_id_basis;
3788 for (sf = table->eviction_fields;
3789 sf < &table->eviction_fields[table->n_eviction_fields];
3792 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
3793 union mf_value value;
3795 mf_get_value(sf->field, &rule->cr.flow, &value);
3797 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
3799 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
3800 unsigned int start = sf->ofs + sf->n_bits;
3801 bitwise_zero(&value, sf->field->n_bytes, start,
3802 sf->field->n_bytes * 8 - start);
3804 hash = hash_bytes(&value, sf->field->n_bytes, hash);
3806 hash = hash_int(hash, 0);
3813 /* Returns an eviction group within 'table' with the given 'id', creating one
3815 static struct eviction_group *
3816 eviction_group_find(struct oftable *table, uint32_t id)
3818 struct eviction_group *evg;
3820 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
3824 evg = xmalloc(sizeof *evg);
3825 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
3826 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
3827 eviction_group_priority(0));
3828 heap_init(&evg->rules);
3833 /* Returns an eviction priority for 'rule'. The return value should be
3834 * interpreted so that higher priorities make a rule more attractive candidates
3837 rule_eviction_priority(struct rule *rule)
3839 long long int hard_expiration;
3840 long long int idle_expiration;
3841 long long int expiration;
3842 uint32_t expiration_offset;
3844 /* Calculate time of expiration. */
3845 hard_expiration = (rule->hard_timeout
3846 ? rule->modified + rule->hard_timeout * 1000
3848 idle_expiration = (rule->idle_timeout
3849 ? rule->used + rule->idle_timeout * 1000
3851 expiration = MIN(hard_expiration, idle_expiration);
3852 if (expiration == LLONG_MAX) {
3856 /* Calculate the time of expiration as a number of (approximate) seconds
3857 * after program startup.
3859 * This should work OK for program runs that last UINT32_MAX seconds or
3860 * less. Therefore, please restart OVS at least once every 136 years. */
3861 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
3863 /* Invert the expiration offset because we're using a max-heap. */
3864 return UINT32_MAX - expiration_offset;
3867 /* Adds 'rule' to an appropriate eviction group for its oftable's
3868 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
3869 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
3872 * The caller must ensure that 'rule' is not already in an eviction group. */
3874 eviction_group_add_rule(struct rule *rule)
3876 struct ofproto *ofproto = rule->ofproto;
3877 struct oftable *table = &ofproto->tables[rule->table_id];
3879 if (table->eviction_fields
3880 && (rule->hard_timeout || rule->idle_timeout)) {
3881 struct eviction_group *evg;
3883 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
3885 rule->eviction_group = evg;
3886 heap_insert(&evg->rules, &rule->evg_node,
3887 rule_eviction_priority(rule));
3888 eviction_group_resized(table, evg);
3894 /* Initializes 'table'. */
3896 oftable_init(struct oftable *table)
3898 memset(table, 0, sizeof *table);
3899 classifier_init(&table->cls);
3902 /* Destroys 'table', including its classifier and eviction groups.
3904 * The caller is responsible for freeing 'table' itself. */
3906 oftable_destroy(struct oftable *table)
3908 assert(classifier_is_empty(&table->cls));
3909 oftable_disable_eviction(table);
3910 classifier_destroy(&table->cls);
3914 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
3915 * string, then 'table' will use its default name.
3917 * This only affects the name exposed for a table exposed through the OpenFlow
3918 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
3920 oftable_set_name(struct oftable *table, const char *name)
3922 if (name && name[0]) {
3923 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
3924 if (!table->name || strncmp(name, table->name, len)) {
3926 table->name = xmemdup0(name, len);
3934 /* oftables support a choice of two policies when adding a rule would cause the
3935 * number of flows in the table to exceed the configured maximum number: either
3936 * they can refuse to add the new flow or they can evict some existing flow.
3937 * This function configures the former policy on 'table'. */
3939 oftable_disable_eviction(struct oftable *table)
3941 if (table->eviction_fields) {
3942 struct eviction_group *evg, *next;
3944 HMAP_FOR_EACH_SAFE (evg, next, id_node,
3945 &table->eviction_groups_by_id) {
3946 eviction_group_destroy(table, evg);
3948 hmap_destroy(&table->eviction_groups_by_id);
3949 heap_destroy(&table->eviction_groups_by_size);
3951 free(table->eviction_fields);
3952 table->eviction_fields = NULL;
3953 table->n_eviction_fields = 0;
3957 /* oftables support a choice of two policies when adding a rule would cause the
3958 * number of flows in the table to exceed the configured maximum number: either
3959 * they can refuse to add the new flow or they can evict some existing flow.
3960 * This function configures the latter policy on 'table', with fairness based
3961 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
3962 * 'n_fields' as 0 disables fairness.) */
3964 oftable_enable_eviction(struct oftable *table,
3965 const struct mf_subfield *fields, size_t n_fields)
3967 struct cls_cursor cursor;
3970 if (table->eviction_fields
3971 && n_fields == table->n_eviction_fields
3973 || !memcmp(fields, table->eviction_fields,
3974 n_fields * sizeof *fields))) {
3979 oftable_disable_eviction(table);
3981 table->n_eviction_fields = n_fields;
3982 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
3984 table->eviction_group_id_basis = random_uint32();
3985 hmap_init(&table->eviction_groups_by_id);
3986 heap_init(&table->eviction_groups_by_size);
3988 cls_cursor_init(&cursor, &table->cls, NULL);
3989 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3990 eviction_group_add_rule(rule);
3994 /* Removes 'rule' from the oftable that contains it. */
3996 oftable_remove_rule(struct rule *rule)
3998 struct ofproto *ofproto = rule->ofproto;
3999 struct oftable *table = &ofproto->tables[rule->table_id];
4001 classifier_remove(&table->cls, &rule->cr);
4002 eviction_group_remove_rule(rule);
4005 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4006 * oftable that has an identical cls_rule. Returns the rule that was removed,
4007 * if any, and otherwise NULL. */
4008 static struct rule *
4009 oftable_replace_rule(struct rule *rule)
4011 struct ofproto *ofproto = rule->ofproto;
4012 struct oftable *table = &ofproto->tables[rule->table_id];
4013 struct rule *victim;
4015 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4017 eviction_group_remove_rule(victim);
4019 eviction_group_add_rule(rule);
4023 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4025 oftable_substitute_rule(struct rule *old, struct rule *new)
4028 oftable_replace_rule(new);
4030 oftable_remove_rule(old);
4034 /* unixctl commands. */
4037 ofproto_lookup(const char *name)
4039 struct ofproto *ofproto;
4041 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4043 if (!strcmp(ofproto->name, name)) {
4051 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4052 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4054 struct ofproto *ofproto;
4058 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4059 ds_put_format(&results, "%s\n", ofproto->name);
4061 unixctl_command_reply(conn, ds_cstr(&results));
4062 ds_destroy(&results);
4066 ofproto_unixctl_init(void)
4068 static bool registered;
4074 unixctl_command_register("ofproto/list", "", 0, 0,
4075 ofproto_unixctl_list, NULL);
4078 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4080 * This is deprecated. It is only for compatibility with broken device drivers
4081 * in old versions of Linux that do not properly support VLANs when VLAN
4082 * devices are not used. When broken device drivers are no longer in
4083 * widespread use, we will delete these interfaces. */
4085 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4086 * (exactly) by an OpenFlow rule in 'ofproto'. */
4088 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4090 const struct oftable *oftable;
4092 free(ofproto->vlan_bitmap);
4093 ofproto->vlan_bitmap = bitmap_allocate(4096);
4094 ofproto->vlans_changed = false;
4096 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4097 const struct cls_table *table;
4099 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4100 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4101 == htons(VLAN_VID_MASK)) {
4102 const struct cls_rule *rule;
4104 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4105 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4106 bitmap_set1(vlan_bitmap, vid);
4107 bitmap_set1(ofproto->vlan_bitmap, vid);
4114 /* Returns true if new VLANs have come into use by the flow table since the
4115 * last call to ofproto_get_vlan_usage().
4117 * We don't track when old VLANs stop being used. */
4119 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4121 return ofproto->vlans_changed;
4124 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4125 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4126 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4127 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4128 * then the VLAN device is un-enslaved. */
4130 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4131 uint16_t realdev_ofp_port, int vid)
4133 struct ofport *ofport;
4136 assert(vlandev_ofp_port != realdev_ofp_port);
4138 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4140 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4141 ofproto->name, vlandev_ofp_port);
4145 if (!ofproto->ofproto_class->set_realdev) {
4146 if (!vlandev_ofp_port) {
4149 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4153 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4155 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4156 ofproto->name, vlandev_ofp_port,
4157 netdev_get_name(ofport->netdev), strerror(error));