2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
37 #include "ofp-print.h"
40 #include "ofproto-provider.h"
41 #include "openflow/nicira-ext.h"
42 #include "openflow/openflow.h"
46 #include "poll-loop.h"
52 #include "unaligned.h"
56 VLOG_DEFINE_THIS_MODULE(ofproto);
58 COVERAGE_DEFINE(ofproto_error);
59 COVERAGE_DEFINE(ofproto_flush);
60 COVERAGE_DEFINE(ofproto_no_packet_in);
61 COVERAGE_DEFINE(ofproto_packet_out);
62 COVERAGE_DEFINE(ofproto_queue_req);
63 COVERAGE_DEFINE(ofproto_recv_openflow);
64 COVERAGE_DEFINE(ofproto_reinit_ports);
65 COVERAGE_DEFINE(ofproto_uninstallable);
66 COVERAGE_DEFINE(ofproto_update_port);
69 S_OPENFLOW, /* Processing OpenFlow commands. */
70 S_EVICT, /* Evicting flows from over-limit tables. */
71 S_FLUSH, /* Deleting all flow table rules. */
74 enum ofoperation_type {
80 /* A single OpenFlow request can execute any number of operations. The
81 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
82 * ofconn to which an error reply should be sent if necessary.
84 * ofproto initiates some operations internally. These operations are still
85 * assigned to groups but will not have an associated ofconn. */
87 struct ofproto *ofproto; /* Owning ofproto. */
88 struct list ofproto_node; /* In ofproto's "pending" list. */
89 struct list ops; /* List of "struct ofoperation"s. */
91 /* Data needed to send OpenFlow reply on failure or to send a buffered
94 * If list_is_empty(ofconn_node) then this ofopgroup never had an
95 * associated ofconn or its ofconn's connection dropped after it initiated
96 * the operation. In the latter case 'ofconn' is a wild pointer that
97 * refers to freed memory, so the 'ofconn' member must be used only if
98 * !list_is_empty(ofconn_node).
100 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
101 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
102 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
103 uint32_t buffer_id; /* Buffer id from original request. */
104 int error; /* 0 if no error yet, otherwise error code. */
107 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
108 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
109 const struct ofp_header *,
111 static void ofopgroup_submit(struct ofopgroup *);
112 static void ofopgroup_destroy(struct ofopgroup *);
114 /* A single flow table operation. */
116 struct ofopgroup *group; /* Owning group. */
117 struct list group_node; /* In ofopgroup's "ops" list. */
118 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
119 struct rule *rule; /* Rule being operated upon. */
120 enum ofoperation_type type; /* Type of operation. */
121 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
122 struct ofpact *ofpacts; /* OFOPERATION_MODIFYING: Replaced actions. */
123 size_t ofpacts_len; /* OFOPERATION_MODIFYING: Bytes of ofpacts. */
124 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
127 static void ofoperation_create(struct ofopgroup *, struct rule *,
128 enum ofoperation_type);
129 static void ofoperation_destroy(struct ofoperation *);
132 static void oftable_init(struct oftable *);
133 static void oftable_destroy(struct oftable *);
135 static void oftable_set_name(struct oftable *, const char *name);
137 static void oftable_disable_eviction(struct oftable *);
138 static void oftable_enable_eviction(struct oftable *,
139 const struct mf_subfield *fields,
142 static void oftable_remove_rule(struct rule *);
143 static struct rule *oftable_replace_rule(struct rule *);
144 static void oftable_substitute_rule(struct rule *old, struct rule *new);
146 /* A set of rules within a single OpenFlow table (oftable) that have the same
147 * values for the oftable's eviction_fields. A rule to be evicted, when one is
148 * needed, is taken from the eviction group that contains the greatest number
151 * An oftable owns any number of eviction groups, each of which contains any
154 * Membership in an eviction group is imprecise, based on the hash of the
155 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
156 * That is, if two rules have different eviction_fields, but those
157 * eviction_fields hash to the same value, then they will belong to the same
158 * eviction_group anyway.
160 * (When eviction is not enabled on an oftable, we don't track any eviction
161 * groups, to save time and space.) */
162 struct eviction_group {
163 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
164 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
165 struct heap rules; /* Contains "struct rule"s. */
168 static struct rule *choose_rule_to_evict(struct oftable *);
169 static void ofproto_evict(struct ofproto *);
170 static uint32_t rule_eviction_priority(struct rule *);
173 static void ofport_destroy__(struct ofport *);
174 static void ofport_destroy(struct ofport *);
176 static void update_port(struct ofproto *, const char *devname);
177 static int init_ports(struct ofproto *);
178 static void reinit_ports(struct ofproto *);
181 static void ofproto_rule_destroy__(struct rule *);
182 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
183 static bool rule_is_modifiable(const struct rule *);
184 static bool rule_is_hidden(const struct rule *);
187 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
188 const struct ofputil_flow_mod *,
189 const struct ofp_header *);
190 static void delete_flow__(struct rule *, struct ofopgroup *);
191 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
192 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
193 const struct ofputil_flow_mod *,
194 const struct ofp_header *);
197 static uint64_t pick_datapath_id(const struct ofproto *);
198 static uint64_t pick_fallback_dpid(void);
199 static void ofproto_destroy__(struct ofproto *);
200 static void update_mtu(struct ofproto *, struct ofport *);
203 static void ofproto_unixctl_init(void);
205 /* All registered ofproto classes, in probe order. */
206 static const struct ofproto_class **ofproto_classes;
207 static size_t n_ofproto_classes;
208 static size_t allocated_ofproto_classes;
210 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
211 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
216 ofproto_initialize(void)
222 ofproto_class_register(&ofproto_dpif_class);
226 /* 'type' should be a normalized datapath type, as returned by
227 * ofproto_normalize_type(). Returns the corresponding ofproto_class
228 * structure, or a null pointer if there is none registered for 'type'. */
229 static const struct ofproto_class *
230 ofproto_class_find__(const char *type)
234 ofproto_initialize();
235 for (i = 0; i < n_ofproto_classes; i++) {
236 const struct ofproto_class *class = ofproto_classes[i];
241 class->enumerate_types(&types);
242 found = sset_contains(&types, type);
243 sset_destroy(&types);
249 VLOG_WARN("unknown datapath type %s", type);
253 /* Registers a new ofproto class. After successful registration, new ofprotos
254 * of that type can be created using ofproto_create(). */
256 ofproto_class_register(const struct ofproto_class *new_class)
260 for (i = 0; i < n_ofproto_classes; i++) {
261 if (ofproto_classes[i] == new_class) {
266 if (n_ofproto_classes >= allocated_ofproto_classes) {
267 ofproto_classes = x2nrealloc(ofproto_classes,
268 &allocated_ofproto_classes,
269 sizeof *ofproto_classes);
271 ofproto_classes[n_ofproto_classes++] = new_class;
275 /* Unregisters a datapath provider. 'type' must have been previously
276 * registered and not currently be in use by any ofprotos. After
277 * unregistration new datapaths of that type cannot be opened using
278 * ofproto_create(). */
280 ofproto_class_unregister(const struct ofproto_class *class)
284 for (i = 0; i < n_ofproto_classes; i++) {
285 if (ofproto_classes[i] == class) {
286 for (i++; i < n_ofproto_classes; i++) {
287 ofproto_classes[i - 1] = ofproto_classes[i];
293 VLOG_WARN("attempted to unregister an ofproto class that is not "
298 /* Clears 'types' and enumerates all registered ofproto types into it. The
299 * caller must first initialize the sset. */
301 ofproto_enumerate_types(struct sset *types)
305 ofproto_initialize();
306 for (i = 0; i < n_ofproto_classes; i++) {
307 ofproto_classes[i]->enumerate_types(types);
311 /* Returns the fully spelled out name for the given ofproto 'type'.
313 * Normalized type string can be compared with strcmp(). Unnormalized type
314 * string might be the same even if they have different spellings. */
316 ofproto_normalize_type(const char *type)
318 return type && type[0] ? type : "system";
321 /* Clears 'names' and enumerates the names of all known created ofprotos with
322 * the given 'type'. The caller must first initialize the sset. Returns 0 if
323 * successful, otherwise a positive errno value.
325 * Some kinds of datapaths might not be practically enumerable. This is not
326 * considered an error. */
328 ofproto_enumerate_names(const char *type, struct sset *names)
330 const struct ofproto_class *class = ofproto_class_find__(type);
331 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
335 ofproto_create(const char *datapath_name, const char *datapath_type,
336 struct ofproto **ofprotop)
338 const struct ofproto_class *class;
339 struct ofproto *ofproto;
344 ofproto_initialize();
345 ofproto_unixctl_init();
347 datapath_type = ofproto_normalize_type(datapath_type);
348 class = ofproto_class_find__(datapath_type);
350 VLOG_WARN("could not create datapath %s of unknown type %s",
351 datapath_name, datapath_type);
355 ofproto = class->alloc();
357 VLOG_ERR("failed to allocate datapath %s of type %s",
358 datapath_name, datapath_type);
363 memset(ofproto, 0, sizeof *ofproto);
364 ofproto->ofproto_class = class;
365 ofproto->name = xstrdup(datapath_name);
366 ofproto->type = xstrdup(datapath_type);
367 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
368 hash_string(ofproto->name, 0));
369 ofproto->datapath_id = 0;
370 ofproto_set_flow_eviction_threshold(ofproto,
371 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
372 ofproto->forward_bpdu = false;
373 ofproto->fallback_dpid = pick_fallback_dpid();
374 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
375 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
376 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
377 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
378 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
379 ofproto->frag_handling = OFPC_FRAG_NORMAL;
380 hmap_init(&ofproto->ports);
381 shash_init(&ofproto->port_by_name);
382 ofproto->tables = NULL;
383 ofproto->n_tables = 0;
384 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
385 ofproto->state = S_OPENFLOW;
386 list_init(&ofproto->pending);
387 ofproto->n_pending = 0;
388 hmap_init(&ofproto->deletions);
389 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
390 ofproto->first_op = ofproto->last_op = LLONG_MIN;
391 ofproto->next_op_report = LLONG_MAX;
392 ofproto->op_backoff = LLONG_MIN;
393 ofproto->vlan_bitmap = NULL;
394 ofproto->vlans_changed = false;
395 ofproto->min_mtu = INT_MAX;
397 error = ofproto->ofproto_class->construct(ofproto);
399 VLOG_ERR("failed to open datapath %s: %s",
400 datapath_name, strerror(error));
401 ofproto_destroy__(ofproto);
405 assert(ofproto->n_tables);
407 ofproto->datapath_id = pick_datapath_id(ofproto);
408 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
416 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
418 struct oftable *table;
420 assert(!ofproto->n_tables);
421 assert(n_tables >= 1 && n_tables <= 255);
423 ofproto->n_tables = n_tables;
424 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
425 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
431 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
433 uint64_t old_dpid = p->datapath_id;
434 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
435 if (p->datapath_id != old_dpid) {
436 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
438 /* Force all active connections to reconnect, since there is no way to
439 * notify a controller that the datapath ID has changed. */
440 ofproto_reconnect_controllers(p);
445 ofproto_set_controllers(struct ofproto *p,
446 const struct ofproto_controller *controllers,
447 size_t n_controllers)
449 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
453 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
455 connmgr_set_fail_mode(p->connmgr, fail_mode);
458 /* Drops the connections between 'ofproto' and all of its controllers, forcing
459 * them to reconnect. */
461 ofproto_reconnect_controllers(struct ofproto *ofproto)
463 connmgr_reconnect(ofproto->connmgr);
466 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
467 * in-band control should guarantee access, in the same way that in-band
468 * control guarantees access to OpenFlow controllers. */
470 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
471 const struct sockaddr_in *extras, size_t n)
473 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
476 /* Sets the OpenFlow queue used by flows set up by in-band control on
477 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
478 * flows will use the default queue. */
480 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
482 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
485 /* Sets the number of flows at which eviction from the kernel flow table
488 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
490 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
491 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
493 ofproto->flow_eviction_threshold = threshold;
497 /* If forward_bpdu is true, the NORMAL action will forward frames with
498 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
499 * the NORMAL action will drop these frames. */
501 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
503 bool old_val = ofproto->forward_bpdu;
504 ofproto->forward_bpdu = forward_bpdu;
505 if (old_val != ofproto->forward_bpdu) {
506 if (ofproto->ofproto_class->forward_bpdu_changed) {
507 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
512 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
513 * 'idle_time', in seconds. */
515 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
517 if (ofproto->ofproto_class->set_mac_idle_time) {
518 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
523 ofproto_set_desc(struct ofproto *p,
524 const char *mfr_desc, const char *hw_desc,
525 const char *sw_desc, const char *serial_desc,
528 struct ofp_desc_stats *ods;
531 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
532 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
533 sizeof ods->mfr_desc);
536 p->mfr_desc = xstrdup(mfr_desc);
539 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
540 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
541 sizeof ods->hw_desc);
544 p->hw_desc = xstrdup(hw_desc);
547 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
548 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
549 sizeof ods->sw_desc);
552 p->sw_desc = xstrdup(sw_desc);
555 if (strlen(serial_desc) >= sizeof ods->serial_num) {
556 VLOG_WARN("truncating serial_desc, must be less than %zu "
558 sizeof ods->serial_num);
560 free(p->serial_desc);
561 p->serial_desc = xstrdup(serial_desc);
564 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
565 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
566 sizeof ods->dp_desc);
569 p->dp_desc = xstrdup(dp_desc);
574 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
576 return connmgr_set_snoops(ofproto->connmgr, snoops);
580 ofproto_set_netflow(struct ofproto *ofproto,
581 const struct netflow_options *nf_options)
583 if (nf_options && sset_is_empty(&nf_options->collectors)) {
587 if (ofproto->ofproto_class->set_netflow) {
588 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
590 return nf_options ? EOPNOTSUPP : 0;
595 ofproto_set_sflow(struct ofproto *ofproto,
596 const struct ofproto_sflow_options *oso)
598 if (oso && sset_is_empty(&oso->targets)) {
602 if (ofproto->ofproto_class->set_sflow) {
603 return ofproto->ofproto_class->set_sflow(ofproto, oso);
605 return oso ? EOPNOTSUPP : 0;
609 /* Spanning Tree Protocol (STP) configuration. */
611 /* Configures STP on 'ofproto' using the settings defined in 's'. If
612 * 's' is NULL, disables STP.
614 * Returns 0 if successful, otherwise a positive errno value. */
616 ofproto_set_stp(struct ofproto *ofproto,
617 const struct ofproto_stp_settings *s)
619 return (ofproto->ofproto_class->set_stp
620 ? ofproto->ofproto_class->set_stp(ofproto, s)
624 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
625 * 'enabled' member of 's' is false, then the other members are not
628 * Returns 0 if successful, otherwise a positive errno value. */
630 ofproto_get_stp_status(struct ofproto *ofproto,
631 struct ofproto_stp_status *s)
633 return (ofproto->ofproto_class->get_stp_status
634 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
638 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
639 * in 's'. The caller is responsible for assigning STP port numbers
640 * (using the 'port_num' member in the range of 1 through 255, inclusive)
641 * and ensuring there are no duplicates. If the 's' is NULL, then STP
642 * is disabled on the port.
644 * Returns 0 if successful, otherwise a positive errno value.*/
646 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
647 const struct ofproto_port_stp_settings *s)
649 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
651 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
652 ofproto->name, ofp_port);
656 return (ofproto->ofproto_class->set_stp_port
657 ? ofproto->ofproto_class->set_stp_port(ofport, s)
661 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
662 * 's'. If the 'enabled' member in 's' is false, then the other members
663 * are not meaningful.
665 * Returns 0 if successful, otherwise a positive errno value.*/
667 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
668 struct ofproto_port_stp_status *s)
670 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
672 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
673 "port %"PRIu16, ofproto->name, ofp_port);
677 return (ofproto->ofproto_class->get_stp_port_status
678 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
682 /* Queue DSCP configuration. */
684 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
685 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
686 * to implement QoS. Instead, it is used to implement features which require
687 * knowledge of what queues exist on a port, and some basic information about
690 * Returns 0 if successful, otherwise a positive errno value. */
692 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
693 const struct ofproto_port_queue *queues,
696 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
699 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
700 ofproto->name, ofp_port);
704 return (ofproto->ofproto_class->set_queues
705 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
709 /* Connectivity Fault Management configuration. */
711 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
713 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
715 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
716 if (ofport && ofproto->ofproto_class->set_cfm) {
717 ofproto->ofproto_class->set_cfm(ofport, NULL);
721 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
722 * basic configuration from the configuration members in 'cfm', and the remote
723 * maintenance point ID from remote_mpid. Ignores the statistics members of
726 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
728 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
729 const struct cfm_settings *s)
731 struct ofport *ofport;
734 ofport = ofproto_get_port(ofproto, ofp_port);
736 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
737 ofproto->name, ofp_port);
741 /* XXX: For configuration simplicity, we only support one remote_mpid
742 * outside of the CFM module. It's not clear if this is the correct long
743 * term solution or not. */
744 error = (ofproto->ofproto_class->set_cfm
745 ? ofproto->ofproto_class->set_cfm(ofport, s)
748 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
749 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
754 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
755 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
756 * 0 if LACP partner information is not current (generally indicating a
757 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
759 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
761 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
762 return (ofport && ofproto->ofproto_class->port_is_lacp_current
763 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
769 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
770 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
771 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
772 * configuration plus, if there is more than one slave, a bonding
775 * If 'aux' is already registered then this function updates its configuration
776 * to 's'. Otherwise, this function registers a new bundle.
778 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
781 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
782 const struct ofproto_bundle_settings *s)
784 return (ofproto->ofproto_class->bundle_set
785 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
789 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
790 * If no such bundle has been registered, this has no effect. */
792 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
794 return ofproto_bundle_register(ofproto, aux, NULL);
798 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
799 * If 'aux' is already registered then this function updates its configuration
800 * to 's'. Otherwise, this function registers a new mirror. */
802 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
803 const struct ofproto_mirror_settings *s)
805 return (ofproto->ofproto_class->mirror_set
806 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
810 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
811 * If no mirror has been registered, this has no effect. */
813 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
815 return ofproto_mirror_register(ofproto, aux, NULL);
818 /* Retrieves statistics from mirror associated with client data pointer
819 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
820 * 'bytes', respectively. If a particular counters is not supported,
821 * the appropriate argument is set to UINT64_MAX. */
823 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
824 uint64_t *packets, uint64_t *bytes)
826 if (!ofproto->ofproto_class->mirror_get_stats) {
827 *packets = *bytes = UINT64_MAX;
831 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
835 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
836 * which all packets are flooded, instead of using MAC learning. If
837 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
839 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
842 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
844 return (ofproto->ofproto_class->set_flood_vlans
845 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
849 /* Returns true if 'aux' is a registered bundle that is currently in use as the
850 * output for a mirror. */
852 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
854 return (ofproto->ofproto_class->is_mirror_output_bundle
855 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
859 /* Configuration of OpenFlow tables. */
861 /* Returns the number of OpenFlow tables in 'ofproto'. */
863 ofproto_get_n_tables(const struct ofproto *ofproto)
865 return ofproto->n_tables;
868 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
869 * settings from 's'. 'table_id' must be in the range 0 through the number of
870 * OpenFlow tables in 'ofproto' minus 1, inclusive.
872 * For read-only tables, only the name may be configured. */
874 ofproto_configure_table(struct ofproto *ofproto, int table_id,
875 const struct ofproto_table_settings *s)
877 struct oftable *table;
879 assert(table_id >= 0 && table_id < ofproto->n_tables);
880 table = &ofproto->tables[table_id];
882 oftable_set_name(table, s->name);
884 if (table->flags & OFTABLE_READONLY) {
889 oftable_enable_eviction(table, s->groups, s->n_groups);
891 oftable_disable_eviction(table);
894 table->max_flows = s->max_flows;
895 if (classifier_count(&table->cls) > table->max_flows
896 && table->eviction_fields) {
897 /* 'table' contains more flows than allowed. We might not be able to
898 * evict them right away because of the asynchronous nature of flow
899 * table changes. Schedule eviction for later. */
900 switch (ofproto->state) {
902 ofproto->state = S_EVICT;
906 /* We're already deleting flows, nothing more to do. */
913 ofproto_has_snoops(const struct ofproto *ofproto)
915 return connmgr_has_snoops(ofproto->connmgr);
919 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
921 connmgr_get_snoops(ofproto->connmgr, snoops);
925 ofproto_flush__(struct ofproto *ofproto)
927 struct ofopgroup *group;
928 struct oftable *table;
930 if (ofproto->ofproto_class->flush) {
931 ofproto->ofproto_class->flush(ofproto);
934 group = ofopgroup_create_unattached(ofproto);
935 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
936 struct rule *rule, *next_rule;
937 struct cls_cursor cursor;
939 if (table->flags & OFTABLE_HIDDEN) {
943 cls_cursor_init(&cursor, &table->cls, NULL);
944 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
945 if (!rule->pending) {
946 ofoperation_create(group, rule, OFOPERATION_DELETE);
947 oftable_remove_rule(rule);
948 ofproto->ofproto_class->rule_destruct(rule);
952 ofopgroup_submit(group);
956 ofproto_destroy__(struct ofproto *ofproto)
958 struct oftable *table;
960 assert(list_is_empty(&ofproto->pending));
961 assert(!ofproto->n_pending);
963 connmgr_destroy(ofproto->connmgr);
965 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
968 free(ofproto->mfr_desc);
969 free(ofproto->hw_desc);
970 free(ofproto->sw_desc);
971 free(ofproto->serial_desc);
972 free(ofproto->dp_desc);
973 hmap_destroy(&ofproto->ports);
974 shash_destroy(&ofproto->port_by_name);
976 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
977 oftable_destroy(table);
979 free(ofproto->tables);
981 hmap_destroy(&ofproto->deletions);
983 free(ofproto->vlan_bitmap);
985 ofproto->ofproto_class->dealloc(ofproto);
989 ofproto_destroy(struct ofproto *p)
991 struct ofport *ofport, *next_ofport;
998 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
999 ofport_destroy(ofport);
1002 p->ofproto_class->destruct(p);
1003 ofproto_destroy__(p);
1006 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1007 * kernel datapath, for example, this destroys the datapath in the kernel, and
1008 * with the netdev-based datapath, it tears down the data structures that
1009 * represent the datapath.
1011 * The datapath should not be currently open as an ofproto. */
1013 ofproto_delete(const char *name, const char *type)
1015 const struct ofproto_class *class = ofproto_class_find__(type);
1016 return (!class ? EAFNOSUPPORT
1017 : !class->del ? EACCES
1018 : class->del(type, name));
1022 process_port_change(struct ofproto *ofproto, int error, char *devname)
1024 if (error == ENOBUFS) {
1025 reinit_ports(ofproto);
1026 } else if (!error) {
1027 update_port(ofproto, devname);
1033 ofproto_run(struct ofproto *p)
1035 struct sset changed_netdevs;
1036 const char *changed_netdev;
1037 struct ofport *ofport;
1040 error = p->ofproto_class->run(p);
1041 if (error && error != EAGAIN) {
1042 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1045 if (p->ofproto_class->port_poll) {
1048 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1049 process_port_change(p, error, devname);
1053 /* Update OpenFlow port status for any port whose netdev has changed.
1055 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1056 * destroyed, so it's not safe to update ports directly from the
1057 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1058 * need this two-phase approach. */
1059 sset_init(&changed_netdevs);
1060 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1061 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1062 if (ofport->change_seq != change_seq) {
1063 ofport->change_seq = change_seq;
1064 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1067 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1068 update_port(p, changed_netdev);
1070 sset_destroy(&changed_netdevs);
1074 connmgr_run(p->connmgr, handle_openflow);
1078 connmgr_run(p->connmgr, NULL);
1080 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1081 p->state = S_OPENFLOW;
1086 connmgr_run(p->connmgr, NULL);
1088 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1089 connmgr_flushed(p->connmgr);
1090 p->state = S_OPENFLOW;
1098 if (time_msec() >= p->next_op_report) {
1099 long long int ago = (time_msec() - p->first_op) / 1000;
1100 long long int interval = (p->last_op - p->first_op) / 1000;
1104 ds_put_format(&s, "%d flow_mods ",
1105 p->n_add + p->n_delete + p->n_modify);
1106 if (interval == ago) {
1107 ds_put_format(&s, "in the last %lld s", ago);
1108 } else if (interval) {
1109 ds_put_format(&s, "in the %lld s starting %lld s ago",
1112 ds_put_format(&s, "%lld s ago", ago);
1115 ds_put_cstr(&s, " (");
1117 ds_put_format(&s, "%d adds, ", p->n_add);
1120 ds_put_format(&s, "%d deletes, ", p->n_delete);
1123 ds_put_format(&s, "%d modifications, ", p->n_modify);
1126 ds_put_char(&s, ')');
1128 VLOG_INFO("%s", ds_cstr(&s));
1131 p->n_add = p->n_delete = p->n_modify = 0;
1132 p->next_op_report = LLONG_MAX;
1138 /* Performs periodic activity required by 'ofproto' that needs to be done
1139 * with the least possible latency.
1141 * It makes sense to call this function a couple of times per poll loop, to
1142 * provide a significant performance boost on some benchmarks with the
1143 * ofproto-dpif implementation. */
1145 ofproto_run_fast(struct ofproto *p)
1149 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1150 if (error && error != EAGAIN) {
1151 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1152 p->name, strerror(error));
1158 ofproto_wait(struct ofproto *p)
1160 struct ofport *ofport;
1162 p->ofproto_class->wait(p);
1163 if (p->ofproto_class->port_poll_wait) {
1164 p->ofproto_class->port_poll_wait(p);
1167 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1168 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1169 poll_immediate_wake();
1175 connmgr_wait(p->connmgr, true);
1180 connmgr_wait(p->connmgr, false);
1181 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1182 poll_immediate_wake();
1189 ofproto_is_alive(const struct ofproto *p)
1191 return connmgr_has_controllers(p->connmgr);
1194 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1195 * memory_report(). */
1197 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1199 const struct oftable *table;
1200 unsigned int n_rules;
1202 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1203 simap_increase(usage, "ops",
1204 ofproto->n_pending + hmap_count(&ofproto->deletions));
1207 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1208 n_rules += classifier_count(&table->cls);
1210 simap_increase(usage, "rules", n_rules);
1212 if (ofproto->ofproto_class->get_memory_usage) {
1213 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1216 connmgr_get_memory_usage(ofproto->connmgr, usage);
1220 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1223 connmgr_get_controller_info(ofproto->connmgr, info);
1227 ofproto_free_ofproto_controller_info(struct shash *info)
1229 connmgr_free_controller_info(info);
1232 /* Makes a deep copy of 'old' into 'port'. */
1234 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1236 port->name = xstrdup(old->name);
1237 port->type = xstrdup(old->type);
1238 port->ofp_port = old->ofp_port;
1241 /* Frees memory allocated to members of 'ofproto_port'.
1243 * Do not call this function on an ofproto_port obtained from
1244 * ofproto_port_dump_next(): that function retains ownership of the data in the
1247 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1249 free(ofproto_port->name);
1250 free(ofproto_port->type);
1253 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1255 * This function provides no status indication. An error status for the entire
1256 * dump operation is provided when it is completed by calling
1257 * ofproto_port_dump_done().
1260 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1261 const struct ofproto *ofproto)
1263 dump->ofproto = ofproto;
1264 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1268 /* Attempts to retrieve another port from 'dump', which must have been created
1269 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1270 * 'port' and returns true. On failure, returns false.
1272 * Failure might indicate an actual error or merely that the last port has been
1273 * dumped. An error status for the entire dump operation is provided when it
1274 * is completed by calling ofproto_port_dump_done().
1276 * The ofproto owns the data stored in 'port'. It will remain valid until at
1277 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1278 * ofproto_port_dump_done(). */
1280 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1281 struct ofproto_port *port)
1283 const struct ofproto *ofproto = dump->ofproto;
1289 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1292 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1298 /* Completes port table dump operation 'dump', which must have been created
1299 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1300 * error-free, otherwise a positive errno value describing the problem. */
1302 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1304 const struct ofproto *ofproto = dump->ofproto;
1306 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1309 return dump->error == EOF ? 0 : dump->error;
1312 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1313 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1314 * is non-null). On failure, returns a positive errno value and sets
1315 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1317 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1318 uint16_t *ofp_portp)
1323 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1325 update_port(ofproto, netdev_get_name(netdev));
1328 *ofp_portp = error ? OFPP_NONE : ofp_port;
1333 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1334 * initializes '*port' appropriately; on failure, returns a positive errno
1337 * The caller owns the data in 'ofproto_port' and must free it with
1338 * ofproto_port_destroy() when it is no longer needed. */
1340 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1341 struct ofproto_port *port)
1345 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1347 memset(port, 0, sizeof *port);
1352 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1353 * Returns 0 if successful, otherwise a positive errno. */
1355 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1357 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1358 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1361 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1362 if (!error && ofport) {
1363 /* 'name' is the netdev's name and update_port() is going to close the
1364 * netdev. Just in case update_port() refers to 'name' after it
1365 * destroys 'ofport', make a copy of it around the update_port()
1367 char *devname = xstrdup(name);
1368 update_port(ofproto, devname);
1374 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1375 * performs the 'n_actions' actions in 'actions'. The new flow will not
1378 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1379 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1380 * controllers; otherwise, it will be hidden.
1382 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1384 * This is a helper function for in-band control and fail-open. */
1386 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1387 const struct ofpact *ofpacts, size_t ofpacts_len)
1389 const struct rule *rule;
1391 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1392 &ofproto->tables[0].cls, cls_rule));
1393 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1394 ofpacts, ofpacts_len)) {
1395 struct ofputil_flow_mod fm;
1397 memset(&fm, 0, sizeof fm);
1399 fm.buffer_id = UINT32_MAX;
1400 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1401 fm.ofpacts_len = ofpacts_len;
1402 add_flow(ofproto, NULL, &fm, NULL);
1407 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1408 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1409 * operation cannot be initiated now but may be retried later.
1411 * This is a helper function for in-band control and fail-open. */
1413 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1415 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1418 /* Searches for a rule with matching criteria exactly equal to 'target' in
1419 * ofproto's table 0 and, if it finds one, deletes it.
1421 * This is a helper function for in-band control and fail-open. */
1423 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1427 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1428 &ofproto->tables[0].cls, target));
1430 /* No such rule -> success. */
1432 } else if (rule->pending) {
1433 /* An operation on the rule is already pending -> failure.
1434 * Caller must retry later if it's important. */
1437 /* Initiate deletion -> success. */
1438 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1439 ofoperation_create(group, rule, OFOPERATION_DELETE);
1440 oftable_remove_rule(rule);
1441 ofproto->ofproto_class->rule_destruct(rule);
1442 ofopgroup_submit(group);
1448 /* Starts the process of deleting all of the flows from all of ofproto's flow
1449 * tables and then reintroducing the flows required by in-band control and
1450 * fail-open. The process will complete in a later call to ofproto_run(). */
1452 ofproto_flush_flows(struct ofproto *ofproto)
1454 COVERAGE_INC(ofproto_flush);
1455 ofproto->state = S_FLUSH;
1459 reinit_ports(struct ofproto *p)
1461 struct ofproto_port_dump dump;
1462 struct sset devnames;
1463 struct ofport *ofport;
1464 struct ofproto_port ofproto_port;
1465 const char *devname;
1467 COVERAGE_INC(ofproto_reinit_ports);
1469 sset_init(&devnames);
1470 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1471 sset_add(&devnames, netdev_get_name(ofport->netdev));
1473 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1474 sset_add(&devnames, ofproto_port.name);
1477 SSET_FOR_EACH (devname, &devnames) {
1478 update_port(p, devname);
1480 sset_destroy(&devnames);
1483 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1484 * netdev cannot be opened. On success, also fills in 'opp'. */
1485 static struct netdev *
1486 ofport_open(const struct ofproto_port *ofproto_port,
1487 struct ofputil_phy_port *pp)
1489 enum netdev_flags flags;
1490 struct netdev *netdev;
1493 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1495 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1496 "cannot be opened (%s)",
1497 ofproto_port->name, ofproto_port->ofp_port,
1498 ofproto_port->name, strerror(error));
1502 pp->port_no = ofproto_port->ofp_port;
1503 netdev_get_etheraddr(netdev, pp->hw_addr);
1504 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1505 netdev_get_flags(netdev, &flags);
1506 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1507 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1508 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1509 &pp->supported, &pp->peer);
1510 pp->curr_speed = netdev_features_to_bps(pp->curr);
1511 pp->max_speed = netdev_features_to_bps(pp->supported);
1516 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1517 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1520 ofport_equal(const struct ofputil_phy_port *a,
1521 const struct ofputil_phy_port *b)
1523 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1524 && a->state == b->state
1525 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1526 && a->curr == b->curr
1527 && a->advertised == b->advertised
1528 && a->supported == b->supported
1529 && a->peer == b->peer
1530 && a->curr_speed == b->curr_speed
1531 && a->max_speed == b->max_speed);
1534 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1535 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1536 * one with the same name or port number). */
1538 ofport_install(struct ofproto *p,
1539 struct netdev *netdev, const struct ofputil_phy_port *pp)
1541 const char *netdev_name = netdev_get_name(netdev);
1542 struct ofport *ofport;
1545 /* Create ofport. */
1546 ofport = p->ofproto_class->port_alloc();
1551 ofport->ofproto = p;
1552 ofport->netdev = netdev;
1553 ofport->change_seq = netdev_change_seq(netdev);
1555 ofport->ofp_port = pp->port_no;
1557 /* Add port to 'p'. */
1558 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1559 shash_add(&p->port_by_name, netdev_name, ofport);
1561 update_mtu(p, ofport);
1563 /* Let the ofproto_class initialize its private data. */
1564 error = p->ofproto_class->port_construct(ofport);
1568 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1572 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1573 p->name, netdev_name, strerror(error));
1575 ofport_destroy__(ofport);
1577 netdev_close(netdev);
1581 /* Removes 'ofport' from 'p' and destroys it. */
1583 ofport_remove(struct ofport *ofport)
1585 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1587 ofport_destroy(ofport);
1590 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1593 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1595 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1597 ofport_remove(port);
1601 /* Updates 'port' with new 'pp' description.
1603 * Does not handle a name or port number change. The caller must implement
1604 * such a change as a delete followed by an add. */
1606 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1608 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1609 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1610 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1611 port->pp.state = pp->state;
1612 port->pp.curr = pp->curr;
1613 port->pp.advertised = pp->advertised;
1614 port->pp.supported = pp->supported;
1615 port->pp.peer = pp->peer;
1616 port->pp.curr_speed = pp->curr_speed;
1617 port->pp.max_speed = pp->max_speed;
1619 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1622 /* Update OpenFlow 'state' in 'port' and notify controller. */
1624 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1626 if (port->pp.state != state) {
1627 port->pp.state = state;
1628 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1634 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1636 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1638 if (port->ofproto->ofproto_class->set_realdev) {
1639 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1641 if (port->ofproto->ofproto_class->set_stp_port) {
1642 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1644 if (port->ofproto->ofproto_class->set_cfm) {
1645 port->ofproto->ofproto_class->set_cfm(port, NULL);
1647 if (port->ofproto->ofproto_class->bundle_remove) {
1648 port->ofproto->ofproto_class->bundle_remove(port);
1654 ofport_destroy__(struct ofport *port)
1656 struct ofproto *ofproto = port->ofproto;
1657 const char *name = netdev_get_name(port->netdev);
1659 hmap_remove(&ofproto->ports, &port->hmap_node);
1660 shash_delete(&ofproto->port_by_name,
1661 shash_find(&ofproto->port_by_name, name));
1663 netdev_close(port->netdev);
1664 ofproto->ofproto_class->port_dealloc(port);
1668 ofport_destroy(struct ofport *port)
1671 port->ofproto->ofproto_class->port_destruct(port);
1672 ofport_destroy__(port);
1677 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1679 struct ofport *port;
1681 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1682 hash_int(ofp_port, 0), &ofproto->ports) {
1683 if (port->ofp_port == ofp_port) {
1691 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1693 struct ofproto *ofproto = port->ofproto;
1696 if (ofproto->ofproto_class->port_get_stats) {
1697 error = ofproto->ofproto_class->port_get_stats(port, stats);
1706 update_port(struct ofproto *ofproto, const char *name)
1708 struct ofproto_port ofproto_port;
1709 struct ofputil_phy_port pp;
1710 struct netdev *netdev;
1711 struct ofport *port;
1713 COVERAGE_INC(ofproto_update_port);
1715 /* Fetch 'name''s location and properties from the datapath. */
1716 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1717 ? ofport_open(&ofproto_port, &pp)
1720 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1721 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1722 struct netdev *old_netdev = port->netdev;
1724 /* 'name' hasn't changed location. Any properties changed? */
1725 if (!ofport_equal(&port->pp, &pp)) {
1726 ofport_modified(port, &pp);
1729 update_mtu(ofproto, port);
1731 /* Install the newly opened netdev in case it has changed.
1732 * Don't close the old netdev yet in case port_modified has to
1733 * remove a retained reference to it.*/
1734 port->netdev = netdev;
1735 port->change_seq = netdev_change_seq(netdev);
1737 if (port->ofproto->ofproto_class->port_modified) {
1738 port->ofproto->ofproto_class->port_modified(port);
1741 netdev_close(old_netdev);
1743 /* If 'port' is nonnull then its name differs from 'name' and thus
1744 * we should delete it. If we think there's a port named 'name'
1745 * then its port number must be wrong now so delete it too. */
1747 ofport_remove(port);
1749 ofport_remove_with_name(ofproto, name);
1750 ofport_install(ofproto, netdev, &pp);
1753 /* Any port named 'name' is gone now. */
1754 ofport_remove_with_name(ofproto, name);
1756 ofproto_port_destroy(&ofproto_port);
1760 init_ports(struct ofproto *p)
1762 struct ofproto_port_dump dump;
1763 struct ofproto_port ofproto_port;
1765 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1766 uint16_t ofp_port = ofproto_port.ofp_port;
1767 if (ofproto_get_port(p, ofp_port)) {
1768 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1770 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1771 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1774 struct ofputil_phy_port pp;
1775 struct netdev *netdev;
1777 netdev = ofport_open(&ofproto_port, &pp);
1779 ofport_install(p, netdev, &pp);
1787 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1788 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1790 find_min_mtu(struct ofproto *p)
1792 struct ofport *ofport;
1795 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1796 struct netdev *netdev = ofport->netdev;
1799 /* Skip any internal ports, since that's what we're trying to
1801 if (!strcmp(netdev_get_type(netdev), "internal")) {
1805 if (netdev_get_mtu(netdev, &dev_mtu)) {
1808 if (!mtu || dev_mtu < mtu) {
1813 return mtu ? mtu: ETH_PAYLOAD_MAX;
1816 /* Update MTU of all datapath devices on 'p' to the minimum of the
1817 * non-datapath ports in event of 'port' added or changed. */
1819 update_mtu(struct ofproto *p, struct ofport *port)
1821 struct ofport *ofport;
1822 struct netdev *netdev = port->netdev;
1823 int dev_mtu, old_min;
1825 if (netdev_get_mtu(netdev, &dev_mtu)) {
1829 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1830 if (dev_mtu > p->min_mtu) {
1831 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1832 dev_mtu = p->min_mtu;
1835 port->mtu = dev_mtu;
1839 /* For non-internal port find new min mtu. */
1840 old_min = p->min_mtu;
1841 port->mtu = dev_mtu;
1842 p->min_mtu = find_min_mtu(p);
1843 if (p->min_mtu == old_min) {
1847 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1848 struct netdev *netdev = ofport->netdev;
1850 if (!strcmp(netdev_get_type(netdev), "internal")) {
1851 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1852 ofport->mtu = p->min_mtu;
1859 ofproto_rule_destroy__(struct rule *rule)
1862 free(rule->ofpacts);
1863 rule->ofproto->ofproto_class->rule_dealloc(rule);
1867 /* This function allows an ofproto implementation to destroy any rules that
1868 * remain when its ->destruct() function is called. The caller must have
1869 * already uninitialized any derived members of 'rule' (step 5 described in the
1870 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1871 * This function implements steps 6 and 7.
1873 * This function should only be called from an ofproto implementation's
1874 * ->destruct() function. It is not suitable elsewhere. */
1876 ofproto_rule_destroy(struct rule *rule)
1878 assert(!rule->pending);
1879 oftable_remove_rule(rule);
1880 ofproto_rule_destroy__(rule);
1883 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1884 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1886 rule_has_out_port(const struct rule *rule, uint16_t port)
1888 return (port == OFPP_NONE
1889 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
1892 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1893 * statistics appropriately. 'packet' must have at least sizeof(struct
1894 * ofp_packet_in) bytes of headroom.
1896 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1897 * with statistics for 'packet' either way.
1899 * Takes ownership of 'packet'. */
1901 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1905 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1907 flow_extract(packet, 0, 0, in_port, &flow);
1908 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1911 /* Returns true if 'rule' should be hidden from the controller.
1913 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1914 * (e.g. by in-band control) and are intentionally hidden from the
1917 rule_is_hidden(const struct rule *rule)
1919 return rule->cr.priority > UINT16_MAX;
1922 static enum oftable_flags
1923 rule_get_flags(const struct rule *rule)
1925 return rule->ofproto->tables[rule->table_id].flags;
1929 rule_is_modifiable(const struct rule *rule)
1931 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1935 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1937 ofconn_send_reply(ofconn, make_echo_reply(oh));
1942 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1944 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1945 struct ofputil_switch_features features;
1946 struct ofport *port;
1950 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1952 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1954 features.datapath_id = ofproto->datapath_id;
1955 features.n_buffers = pktbuf_capacity();
1956 features.n_tables = ofproto->n_tables;
1957 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1958 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1960 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
1963 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
1965 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1966 ofputil_put_switch_features_port(&port->pp, b);
1969 ofconn_send_reply(ofconn, b);
1974 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1976 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1977 struct ofp_switch_config *osc;
1978 enum ofp_config_flags flags;
1982 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1983 flags = ofproto->frag_handling;
1984 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1985 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1987 osc->flags = htons(flags);
1988 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1989 ofconn_send_reply(ofconn, buf);
1995 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1997 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1998 uint16_t flags = ntohs(osc->flags);
2000 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2001 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2002 enum ofp_config_flags cur = ofproto->frag_handling;
2003 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2005 assert((cur & OFPC_FRAG_MASK) == cur);
2007 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2008 ofproto->frag_handling = next;
2010 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2012 ofputil_frag_handling_to_string(next));
2016 ofconn_set_invalid_ttl_to_controller(ofconn,
2017 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2019 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2024 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2025 * error message code for the caller to propagate upward. Otherwise, returns
2028 * The log message mentions 'msg_type'. */
2030 reject_slave_controller(struct ofconn *ofconn)
2032 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2033 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2034 return OFPERR_OFPBRC_EPERM;
2041 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
2043 struct ofproto *p = ofconn_get_ofproto(ofconn);
2044 struct ofputil_packet_out po;
2045 struct ofpbuf *payload;
2046 uint64_t ofpacts_stub[1024 / 8];
2047 struct ofpbuf ofpacts;
2051 COVERAGE_INC(ofproto_packet_out);
2053 error = reject_slave_controller(ofconn);
2058 /* Decode message. */
2059 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2060 error = ofputil_decode_packet_out(&po, opo, &ofpacts);
2062 goto exit_free_ofpacts;
2066 if (po.buffer_id != UINT32_MAX) {
2067 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2068 if (error || !payload) {
2069 goto exit_free_ofpacts;
2072 payload = xmalloc(sizeof *payload);
2073 ofpbuf_use_const(payload, po.packet, po.packet_len);
2076 /* Send out packet. */
2077 flow_extract(payload, 0, 0, po.in_port, &flow);
2078 error = p->ofproto_class->packet_out(p, payload, &flow,
2079 po.ofpacts, po.ofpacts_len);
2080 ofpbuf_delete(payload);
2083 ofpbuf_uninit(&ofpacts);
2089 update_port_config(struct ofport *port,
2090 enum ofputil_port_config config,
2091 enum ofputil_port_config mask)
2093 enum ofputil_port_config old_config = port->pp.config;
2094 enum ofputil_port_config toggle;
2096 toggle = (config ^ port->pp.config) & mask;
2097 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2098 if (config & OFPUTIL_PC_PORT_DOWN) {
2099 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2101 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2103 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2106 port->pp.config ^= toggle;
2107 if (port->pp.config != old_config) {
2108 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2113 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2115 struct ofproto *p = ofconn_get_ofproto(ofconn);
2116 struct ofputil_port_mod pm;
2117 struct ofport *port;
2120 error = reject_slave_controller(ofconn);
2125 error = ofputil_decode_port_mod(oh, &pm);
2130 port = ofproto_get_port(p, pm.port_no);
2132 return OFPERR_OFPPMFC_BAD_PORT;
2133 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2134 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2136 update_port_config(port, pm.config, pm.mask);
2138 netdev_set_advertisements(port->netdev, pm.advertise);
2145 handle_desc_stats_request(struct ofconn *ofconn,
2146 const struct ofp_stats_msg *request)
2148 struct ofproto *p = ofconn_get_ofproto(ofconn);
2149 struct ofp_desc_stats *ods;
2152 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2153 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2154 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2155 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2156 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2157 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2158 ofconn_send_reply(ofconn, msg);
2164 handle_table_stats_request(struct ofconn *ofconn,
2165 const struct ofp_stats_msg *request)
2167 struct ofproto *p = ofconn_get_ofproto(ofconn);
2168 struct ofp_table_stats *ots;
2172 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2174 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2175 for (i = 0; i < p->n_tables; i++) {
2176 ots[i].table_id = i;
2177 sprintf(ots[i].name, "table%zu", i);
2178 ots[i].wildcards = htonl(OFPFW10_ALL);
2179 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2180 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2183 p->ofproto_class->get_tables(p, ots);
2185 for (i = 0; i < p->n_tables; i++) {
2186 const struct oftable *table = &p->tables[i];
2189 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2192 if (table->max_flows < ntohl(ots[i].max_entries)) {
2193 ots[i].max_entries = htonl(table->max_flows);
2197 ofconn_send_reply(ofconn, msg);
2202 append_port_stat(struct ofport *port, struct list *replies)
2204 struct netdev_stats stats;
2205 struct ofp_port_stats *ops;
2207 /* Intentionally ignore return value, since errors will set
2208 * 'stats' to all-1s, which is correct for OpenFlow, and
2209 * netdev_get_stats() will log errors. */
2210 ofproto_port_get_stats(port, &stats);
2212 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2213 ops->port_no = htons(port->pp.port_no);
2214 memset(ops->pad, 0, sizeof ops->pad);
2215 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2216 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2217 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2218 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2219 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2220 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2221 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2222 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2223 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2224 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2225 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2226 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2230 handle_port_stats_request(struct ofconn *ofconn,
2231 const struct ofp_port_stats_request *psr)
2233 struct ofproto *p = ofconn_get_ofproto(ofconn);
2234 struct ofport *port;
2235 struct list replies;
2237 ofputil_start_stats_reply(&psr->osm, &replies);
2238 if (psr->port_no != htons(OFPP_NONE)) {
2239 port = ofproto_get_port(p, ntohs(psr->port_no));
2241 append_port_stat(port, &replies);
2244 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2245 append_port_stat(port, &replies);
2249 ofconn_send_replies(ofconn, &replies);
2254 handle_port_desc_stats_request(struct ofconn *ofconn,
2255 const struct ofp_stats_msg *osm)
2257 struct ofproto *p = ofconn_get_ofproto(ofconn);
2258 struct ofport *port;
2259 struct list replies;
2261 ofputil_start_stats_reply(osm, &replies);
2263 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2264 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2265 &port->pp, &replies);
2268 ofconn_send_replies(ofconn, &replies);
2273 calc_flow_duration__(long long int start, long long int now,
2274 uint32_t *sec, uint32_t *nsec)
2276 long long int msecs = now - start;
2277 *sec = msecs / 1000;
2278 *nsec = (msecs % 1000) * (1000 * 1000);
2281 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2282 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2284 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2286 return (table_id == 0xff || table_id < ofproto->n_tables
2288 : OFPERR_NXBRC_BAD_TABLE_ID);
2292 static struct oftable *
2293 next_visible_table(struct ofproto *ofproto, uint8_t table_id)
2295 struct oftable *table;
2297 for (table = &ofproto->tables[table_id];
2298 table < &ofproto->tables[ofproto->n_tables];
2300 if (!(table->flags & OFTABLE_HIDDEN)) {
2308 static struct oftable *
2309 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2311 if (table_id == 0xff) {
2312 return next_visible_table(ofproto, 0);
2313 } else if (table_id < ofproto->n_tables) {
2314 return &ofproto->tables[table_id];
2320 static struct oftable *
2321 next_matching_table(struct ofproto *ofproto,
2322 struct oftable *table, uint8_t table_id)
2324 return (table_id == 0xff
2325 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2329 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2331 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2332 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2334 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2335 * only once, for that table. (This can be used to access tables marked
2338 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2339 * entered at all. (Perhaps you should have validated TABLE_ID with
2340 * check_table_id().)
2342 * All parameters are evaluated multiple times.
2344 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2345 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2347 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2349 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2350 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2351 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2354 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2355 * to 'out_port' are included.
2357 * Hidden rules are always omitted.
2359 * Returns 0 on success, otherwise an OpenFlow error code. */
2361 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2362 const struct cls_rule *match,
2363 ovs_be64 cookie, ovs_be64 cookie_mask,
2364 uint16_t out_port, struct list *rules)
2366 struct oftable *table;
2369 error = check_table_id(ofproto, table_id);
2375 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2376 struct cls_cursor cursor;
2379 cls_cursor_init(&cursor, &table->cls, match);
2380 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2381 if (rule->pending) {
2382 return OFPROTO_POSTPONE;
2384 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2385 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2386 list_push_back(rules, &rule->ofproto_node);
2393 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2394 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2395 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2398 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2399 * to 'out_port' are included.
2401 * Hidden rules are always omitted.
2403 * Returns 0 on success, otherwise an OpenFlow error code. */
2405 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2406 const struct cls_rule *match,
2407 ovs_be64 cookie, ovs_be64 cookie_mask,
2408 uint16_t out_port, struct list *rules)
2410 struct oftable *table;
2413 error = check_table_id(ofproto, table_id);
2419 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2422 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2425 if (rule->pending) {
2426 return OFPROTO_POSTPONE;
2428 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2429 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2430 list_push_back(rules, &rule->ofproto_node);
2437 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2438 * forced into the range of a uint16_t. */
2440 age_secs(long long int age_ms)
2442 return (age_ms < 0 ? 0
2443 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2444 : (unsigned int) age_ms / 1000);
2448 handle_flow_stats_request(struct ofconn *ofconn,
2449 const struct ofp_stats_msg *osm)
2451 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2452 struct ofputil_flow_stats_request fsr;
2453 struct list replies;
2458 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2463 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2464 fsr.cookie, fsr.cookie_mask,
2465 fsr.out_port, &rules);
2470 ofputil_start_stats_reply(osm, &replies);
2471 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2472 long long int now = time_msec();
2473 struct ofputil_flow_stats fs;
2476 fs.cookie = rule->flow_cookie;
2477 fs.table_id = rule->table_id;
2478 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2480 fs.idle_timeout = rule->idle_timeout;
2481 fs.hard_timeout = rule->hard_timeout;
2482 fs.idle_age = age_secs(now - rule->used);
2483 fs.hard_age = age_secs(now - rule->modified);
2484 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2486 fs.ofpacts = rule->ofpacts;
2487 fs.ofpacts_len = rule->ofpacts_len;
2488 ofputil_append_flow_stats_reply(&fs, &replies);
2490 ofconn_send_replies(ofconn, &replies);
2496 flow_stats_ds(struct rule *rule, struct ds *results)
2498 uint64_t packet_count, byte_count;
2500 rule->ofproto->ofproto_class->rule_get_stats(rule,
2501 &packet_count, &byte_count);
2503 if (rule->table_id != 0) {
2504 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2506 ds_put_format(results, "duration=%llds, ",
2507 (time_msec() - rule->created) / 1000);
2508 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2509 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2510 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2511 cls_rule_format(&rule->cr, results);
2512 ds_put_char(results, ',');
2513 if (rule->ofpacts_len > 0) {
2514 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2516 ds_put_cstr(results, "drop");
2518 ds_put_cstr(results, "\n");
2521 /* Adds a pretty-printed description of all flows to 'results', including
2522 * hidden flows (e.g., set up by in-band control). */
2524 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2526 struct oftable *table;
2528 OFPROTO_FOR_EACH_TABLE (table, p) {
2529 struct cls_cursor cursor;
2532 cls_cursor_init(&cursor, &table->cls, NULL);
2533 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2534 flow_stats_ds(rule, results);
2539 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2540 * '*engine_type' and '*engine_id', respectively. */
2542 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2543 uint8_t *engine_type, uint8_t *engine_id)
2545 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2548 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2549 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2550 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2551 * and -1 if CFM is not enabled on 'port'. */
2553 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2555 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2556 return (ofport && ofproto->ofproto_class->get_cfm_fault
2557 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2561 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2562 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2563 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2564 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2566 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2567 uint16_t ofp_port, const uint64_t **rmps,
2570 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2574 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2575 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2580 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2581 * integer value between 0 and 100 to indicate the health of the port as a
2582 * percentage which is the average of cfm health of all the remote_mpids or
2583 * returns -1 if CFM is not enabled on 'ofport'. */
2585 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2587 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2588 return (ofport && ofproto->ofproto_class->get_cfm_health
2589 ? ofproto->ofproto_class->get_cfm_health(ofport)
2594 handle_aggregate_stats_request(struct ofconn *ofconn,
2595 const struct ofp_stats_msg *osm)
2597 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2598 struct ofputil_flow_stats_request request;
2599 struct ofputil_aggregate_stats stats;
2600 bool unknown_packets, unknown_bytes;
2601 struct ofpbuf *reply;
2606 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2611 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2612 request.cookie, request.cookie_mask,
2613 request.out_port, &rules);
2618 memset(&stats, 0, sizeof stats);
2619 unknown_packets = unknown_bytes = false;
2620 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2621 uint64_t packet_count;
2622 uint64_t byte_count;
2624 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2627 if (packet_count == UINT64_MAX) {
2628 unknown_packets = true;
2630 stats.packet_count += packet_count;
2633 if (byte_count == UINT64_MAX) {
2634 unknown_bytes = true;
2636 stats.byte_count += byte_count;
2641 if (unknown_packets) {
2642 stats.packet_count = UINT64_MAX;
2644 if (unknown_bytes) {
2645 stats.byte_count = UINT64_MAX;
2648 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2649 ofconn_send_reply(ofconn, reply);
2654 struct queue_stats_cbdata {
2655 struct ofport *ofport;
2656 struct list replies;
2660 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2661 const struct netdev_queue_stats *stats)
2663 struct ofp_queue_stats *reply;
2665 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2666 reply->port_no = htons(cbdata->ofport->pp.port_no);
2667 memset(reply->pad, 0, sizeof reply->pad);
2668 reply->queue_id = htonl(queue_id);
2669 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2670 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2671 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2675 handle_queue_stats_dump_cb(uint32_t queue_id,
2676 struct netdev_queue_stats *stats,
2679 struct queue_stats_cbdata *cbdata = cbdata_;
2681 put_queue_stats(cbdata, queue_id, stats);
2685 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2686 struct queue_stats_cbdata *cbdata)
2688 cbdata->ofport = port;
2689 if (queue_id == OFPQ_ALL) {
2690 netdev_dump_queue_stats(port->netdev,
2691 handle_queue_stats_dump_cb, cbdata);
2693 struct netdev_queue_stats stats;
2695 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2696 put_queue_stats(cbdata, queue_id, &stats);
2698 return OFPERR_OFPQOFC_BAD_QUEUE;
2705 handle_queue_stats_request(struct ofconn *ofconn,
2706 const struct ofp_queue_stats_request *qsr)
2708 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2709 struct queue_stats_cbdata cbdata;
2710 unsigned int port_no;
2711 struct ofport *port;
2715 COVERAGE_INC(ofproto_queue_req);
2717 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2719 port_no = ntohs(qsr->port_no);
2720 queue_id = ntohl(qsr->queue_id);
2721 if (port_no == OFPP_ALL) {
2722 error = OFPERR_OFPQOFC_BAD_QUEUE;
2723 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2724 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2729 port = ofproto_get_port(ofproto, port_no);
2731 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2732 : OFPERR_OFPQOFC_BAD_PORT);
2735 ofconn_send_replies(ofconn, &cbdata.replies);
2737 ofpbuf_list_delete(&cbdata.replies);
2744 is_flow_deletion_pending(const struct ofproto *ofproto,
2745 const struct cls_rule *cls_rule,
2748 if (!hmap_is_empty(&ofproto->deletions)) {
2749 struct ofoperation *op;
2751 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2752 cls_rule_hash(cls_rule, table_id),
2753 &ofproto->deletions) {
2754 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2763 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2764 * in which no matching flow already exists in the flow table.
2766 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2767 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2768 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2769 * initiated now but may be retried later.
2771 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2772 * ownership remains with the caller.
2774 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2777 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2778 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2780 struct oftable *table;
2781 struct ofopgroup *group;
2782 struct rule *victim;
2786 error = check_table_id(ofproto, fm->table_id);
2792 if (fm->table_id == 0xff) {
2794 if (ofproto->ofproto_class->rule_choose_table) {
2795 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2800 assert(table_id < ofproto->n_tables);
2801 table = &ofproto->tables[table_id];
2803 table = &ofproto->tables[0];
2805 } else if (fm->table_id < ofproto->n_tables) {
2806 table = &ofproto->tables[fm->table_id];
2808 return OFPERR_NXFMFC_BAD_TABLE_ID;
2811 if (table->flags & OFTABLE_READONLY) {
2812 return OFPERR_OFPBRC_EPERM;
2815 /* Check for overlap, if requested. */
2816 if (fm->flags & OFPFF_CHECK_OVERLAP
2817 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2818 return OFPERR_OFPFMFC_OVERLAP;
2821 /* Serialize against pending deletion. */
2822 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2823 return OFPROTO_POSTPONE;
2826 /* Allocate new rule. */
2827 rule = ofproto->ofproto_class->rule_alloc();
2829 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2830 ofproto->name, strerror(error));
2833 rule->ofproto = ofproto;
2835 rule->pending = NULL;
2836 rule->flow_cookie = fm->new_cookie;
2837 rule->created = rule->modified = rule->used = time_msec();
2838 rule->idle_timeout = fm->idle_timeout;
2839 rule->hard_timeout = fm->hard_timeout;
2840 rule->table_id = table - ofproto->tables;
2841 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2842 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2843 rule->ofpacts_len = fm->ofpacts_len;
2844 rule->evictable = true;
2845 rule->eviction_group = NULL;
2847 /* Insert new rule. */
2848 victim = oftable_replace_rule(rule);
2849 if (victim && !rule_is_modifiable(victim)) {
2850 error = OFPERR_OFPBRC_EPERM;
2851 } else if (victim && victim->pending) {
2852 error = OFPROTO_POSTPONE;
2856 if (classifier_count(&table->cls) > table->max_flows) {
2859 was_evictable = rule->evictable;
2860 rule->evictable = false;
2861 evict = choose_rule_to_evict(table);
2862 rule->evictable = was_evictable;
2865 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2867 } else if (evict->pending) {
2868 error = OFPROTO_POSTPONE;
2875 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2876 ofoperation_create(group, rule, OFOPERATION_ADD);
2877 rule->pending->victim = victim;
2879 error = ofproto->ofproto_class->rule_construct(rule);
2881 ofoperation_destroy(rule->pending);
2883 delete_flow__(evict, group);
2885 ofopgroup_submit(group);
2889 /* Back out if an error occurred. */
2891 oftable_substitute_rule(rule, victim);
2892 ofproto_rule_destroy__(rule);
2897 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2899 /* Modifies the rules listed in 'rules', changing their actions to match those
2902 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2905 * Returns 0 on success, otherwise an OpenFlow error code. */
2907 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2908 const struct ofputil_flow_mod *fm,
2909 const struct ofp_header *request, struct list *rules)
2911 struct ofopgroup *group;
2915 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2916 error = OFPERR_OFPBRC_EPERM;
2917 LIST_FOR_EACH (rule, ofproto_node, rules) {
2918 if (rule_is_modifiable(rule)) {
2919 /* At least one rule is modifiable, don't report EPERM error. */
2925 if (!ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2926 rule->ofpacts, rule->ofpacts_len)) {
2927 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2928 rule->pending->ofpacts = rule->ofpacts;
2929 rule->pending->ofpacts_len = rule->ofpacts_len;
2930 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2931 rule->ofpacts_len = fm->ofpacts_len;
2932 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2934 rule->modified = time_msec();
2936 if (fm->new_cookie != htonll(UINT64_MAX)) {
2937 rule->flow_cookie = fm->new_cookie;
2940 ofopgroup_submit(group);
2945 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2948 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2951 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2952 const struct ofputil_flow_mod *fm,
2953 const struct ofp_header *request)
2958 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2959 fm->cookie, fm->cookie_mask,
2963 } else if (list_is_empty(&rules)) {
2964 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2966 return modify_flows__(ofproto, ofconn, fm, request, &rules);
2970 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2973 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2976 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2977 const struct ofputil_flow_mod *fm,
2978 const struct ofp_header *request)
2983 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2984 fm->cookie, fm->cookie_mask,
2989 } else if (list_is_empty(&rules)) {
2990 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2992 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2993 fm, request, &rules)
2998 /* OFPFC_DELETE implementation. */
3001 delete_flow__(struct rule *rule, struct ofopgroup *group)
3003 struct ofproto *ofproto = rule->ofproto;
3005 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3007 ofoperation_create(group, rule, OFOPERATION_DELETE);
3008 oftable_remove_rule(rule);
3009 ofproto->ofproto_class->rule_destruct(rule);
3012 /* Deletes the rules listed in 'rules'.
3014 * Returns 0 on success, otherwise an OpenFlow error code. */
3016 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3017 const struct ofp_header *request, struct list *rules)
3019 struct rule *rule, *next;
3020 struct ofopgroup *group;
3022 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3023 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3024 delete_flow__(rule, group);
3026 ofopgroup_submit(group);
3031 /* Implements OFPFC_DELETE. */
3033 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3034 const struct ofputil_flow_mod *fm,
3035 const struct ofp_header *request)
3040 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3041 fm->cookie, fm->cookie_mask,
3042 fm->out_port, &rules);
3043 return (error ? error
3044 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3049 /* Implements OFPFC_DELETE_STRICT. */
3051 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3052 const struct ofputil_flow_mod *fm,
3053 const struct ofp_header *request)
3058 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3059 fm->cookie, fm->cookie_mask,
3060 fm->out_port, &rules);
3061 return (error ? error
3062 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3068 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3070 struct ofputil_flow_removed fr;
3072 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
3077 fr.cookie = rule->flow_cookie;
3079 calc_flow_duration__(rule->created, time_msec(),
3080 &fr.duration_sec, &fr.duration_nsec);
3081 fr.idle_timeout = rule->idle_timeout;
3082 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3085 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3089 ofproto_rule_update_used(struct rule *rule, long long int used)
3091 if (used > rule->used) {
3092 struct eviction_group *evg = rule->eviction_group;
3096 heap_change(&evg->rules, &rule->evg_node,
3097 rule_eviction_priority(rule));
3102 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3103 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3106 * ofproto implementation ->run() functions should use this function to expire
3107 * OpenFlow flows. */
3109 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3111 struct ofproto *ofproto = rule->ofproto;
3112 struct ofopgroup *group;
3114 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3116 ofproto_rule_send_removed(rule, reason);
3118 group = ofopgroup_create_unattached(ofproto);
3119 ofoperation_create(group, rule, OFOPERATION_DELETE);
3120 oftable_remove_rule(rule);
3121 ofproto->ofproto_class->rule_destruct(rule);
3122 ofopgroup_submit(group);
3126 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3128 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3129 struct ofputil_flow_mod fm;
3130 uint64_t ofpacts_stub[1024 / 8];
3131 struct ofpbuf ofpacts;
3135 error = reject_slave_controller(ofconn);
3140 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3141 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3144 goto exit_free_ofpacts;
3147 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3148 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3149 if (fm.flags & OFPFF_EMERG) {
3150 /* We do not support the emergency flow cache. It will hopefully get
3151 * dropped from OpenFlow in the near future. There is no good error
3152 * code, so just state that the flow table is full. */
3153 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
3155 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3158 goto exit_free_ofpacts;
3161 /* Record the operation for logging a summary report. */
3162 switch (fm.command) {
3168 case OFPFC_MODIFY_STRICT:
3169 ofproto->n_modify++;
3173 case OFPFC_DELETE_STRICT:
3174 ofproto->n_delete++;
3179 if (ofproto->next_op_report == LLONG_MAX) {
3180 ofproto->first_op = now;
3181 ofproto->next_op_report = MAX(now + 10 * 1000,
3182 ofproto->op_backoff);
3183 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3185 ofproto->last_op = now;
3188 ofpbuf_uninit(&ofpacts);
3194 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3195 const struct ofputil_flow_mod *fm,
3196 const struct ofp_header *oh)
3198 if (ofproto->n_pending >= 50) {
3199 assert(!list_is_empty(&ofproto->pending));
3200 return OFPROTO_POSTPONE;
3203 switch (fm->command) {
3205 return add_flow(ofproto, ofconn, fm, oh);
3208 return modify_flows_loose(ofproto, ofconn, fm, oh);
3210 case OFPFC_MODIFY_STRICT:
3211 return modify_flow_strict(ofproto, ofconn, fm, oh);
3214 return delete_flows_loose(ofproto, ofconn, fm, oh);
3216 case OFPFC_DELETE_STRICT:
3217 return delete_flow_strict(ofproto, ofconn, fm, oh);
3220 if (fm->command > 0xff) {
3221 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
3222 "flow_mod_table_id extension is not enabled");
3224 return OFPERR_OFPFMFC_BAD_COMMAND;
3229 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3231 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3232 struct nx_role_request *reply;
3236 role = ntohl(nrr->role);
3237 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3238 && role != NX_ROLE_SLAVE) {
3239 return OFPERR_OFPRRFC_BAD_ROLE;
3242 if (ofconn_get_role(ofconn) != role
3243 && ofconn_has_pending_opgroups(ofconn)) {
3244 return OFPROTO_POSTPONE;
3247 ofconn_set_role(ofconn, role);
3249 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3250 reply->role = htonl(role);
3251 ofconn_send_reply(ofconn, buf);
3257 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3258 const struct ofp_header *oh)
3260 const struct nx_flow_mod_table_id *msg
3261 = (const struct nx_flow_mod_table_id *) oh;
3262 enum ofputil_protocol cur, next;
3264 cur = ofconn_get_protocol(ofconn);
3265 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3266 ofconn_set_protocol(ofconn, next);
3272 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3274 const struct nx_set_flow_format *msg
3275 = (const struct nx_set_flow_format *) oh;
3276 enum ofputil_protocol cur, next;
3277 enum ofputil_protocol next_base;
3279 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3281 return OFPERR_OFPBRC_EPERM;
3284 cur = ofconn_get_protocol(ofconn);
3285 next = ofputil_protocol_set_base(cur, next_base);
3286 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3287 /* Avoid sending async messages in surprising protocol. */
3288 return OFPROTO_POSTPONE;
3291 ofconn_set_protocol(ofconn, next);
3296 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3297 const struct ofp_header *oh)
3299 const struct nx_set_packet_in_format *msg;
3302 msg = (const struct nx_set_packet_in_format *) oh;
3303 format = ntohl(msg->format);
3304 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3305 return OFPERR_OFPBRC_EPERM;
3308 if (format != ofconn_get_packet_in_format(ofconn)
3309 && ofconn_has_pending_opgroups(ofconn)) {
3310 /* Avoid sending async message in surprsing packet in format. */
3311 return OFPROTO_POSTPONE;
3314 ofconn_set_packet_in_format(ofconn, format);
3319 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3321 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3322 uint32_t master[OAM_N_TYPES];
3323 uint32_t slave[OAM_N_TYPES];
3325 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3326 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3327 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3329 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3330 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3331 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3333 ofconn_set_async_config(ofconn, master, slave);
3334 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3335 !ofconn_get_miss_send_len(ofconn)) {
3336 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3343 handle_nxt_set_controller_id(struct ofconn *ofconn,
3344 const struct ofp_header *oh)
3346 const struct nx_controller_id *nci;
3348 nci = (const struct nx_controller_id *) oh;
3349 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3350 return OFPERR_NXBRC_MUST_BE_ZERO;
3353 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3358 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3362 if (ofconn_has_pending_opgroups(ofconn)) {
3363 return OFPROTO_POSTPONE;
3366 make_openflow_xid(sizeof *oh, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3367 ofconn_send_reply(ofconn, buf);
3372 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3374 const struct ofp_header *oh = msg->data;
3375 const struct ofputil_msg_type *type;
3378 error = ofputil_decode_msg_type(oh, &type);
3383 switch (ofputil_msg_type_code(type)) {
3384 /* OpenFlow requests. */
3385 case OFPUTIL_OFPT_ECHO_REQUEST:
3386 return handle_echo_request(ofconn, oh);
3388 case OFPUTIL_OFPT_FEATURES_REQUEST:
3389 return handle_features_request(ofconn, oh);
3391 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3392 return handle_get_config_request(ofconn, oh);
3394 case OFPUTIL_OFPT_SET_CONFIG:
3395 return handle_set_config(ofconn, msg->data);
3397 case OFPUTIL_OFPT_PACKET_OUT:
3398 return handle_packet_out(ofconn, msg->data);
3400 case OFPUTIL_OFPT_PORT_MOD:
3401 return handle_port_mod(ofconn, oh);
3403 case OFPUTIL_OFPT_FLOW_MOD:
3404 return handle_flow_mod(ofconn, oh);
3406 case OFPUTIL_OFPT_BARRIER_REQUEST:
3407 return handle_barrier_request(ofconn, oh);
3409 /* OpenFlow replies. */
3410 case OFPUTIL_OFPT_ECHO_REPLY:
3413 /* Nicira extension requests. */
3414 case OFPUTIL_NXT_ROLE_REQUEST:
3415 return handle_role_request(ofconn, oh);
3417 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3418 return handle_nxt_flow_mod_table_id(ofconn, oh);
3420 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3421 return handle_nxt_set_flow_format(ofconn, oh);
3423 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3424 return handle_nxt_set_packet_in_format(ofconn, oh);
3426 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3427 return handle_nxt_set_controller_id(ofconn, oh);
3429 case OFPUTIL_NXT_FLOW_MOD:
3430 return handle_flow_mod(ofconn, oh);
3432 case OFPUTIL_NXT_FLOW_AGE:
3433 /* Nothing to do. */
3436 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3437 return handle_nxt_set_async_config(ofconn, oh);
3439 /* Statistics requests. */
3440 case OFPUTIL_OFPST_DESC_REQUEST:
3441 return handle_desc_stats_request(ofconn, msg->data);
3443 case OFPUTIL_OFPST_FLOW_REQUEST:
3444 case OFPUTIL_NXST_FLOW_REQUEST:
3445 return handle_flow_stats_request(ofconn, msg->data);
3447 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3448 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3449 return handle_aggregate_stats_request(ofconn, msg->data);
3451 case OFPUTIL_OFPST_TABLE_REQUEST:
3452 return handle_table_stats_request(ofconn, msg->data);
3454 case OFPUTIL_OFPST_PORT_REQUEST:
3455 return handle_port_stats_request(ofconn, msg->data);
3457 case OFPUTIL_OFPST_QUEUE_REQUEST:
3458 return handle_queue_stats_request(ofconn, msg->data);
3460 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3461 return handle_port_desc_stats_request(ofconn, msg->data);
3463 case OFPUTIL_MSG_INVALID:
3464 case OFPUTIL_OFPT_HELLO:
3465 case OFPUTIL_OFPT_ERROR:
3466 case OFPUTIL_OFPT_FEATURES_REPLY:
3467 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3468 case OFPUTIL_OFPT_PACKET_IN:
3469 case OFPUTIL_OFPT_FLOW_REMOVED:
3470 case OFPUTIL_OFPT_PORT_STATUS:
3471 case OFPUTIL_OFPT_BARRIER_REPLY:
3472 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3473 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3474 case OFPUTIL_OFPST_DESC_REPLY:
3475 case OFPUTIL_OFPST_FLOW_REPLY:
3476 case OFPUTIL_OFPST_QUEUE_REPLY:
3477 case OFPUTIL_OFPST_PORT_REPLY:
3478 case OFPUTIL_OFPST_TABLE_REPLY:
3479 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3480 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3481 case OFPUTIL_NXT_ROLE_REPLY:
3482 case OFPUTIL_NXT_FLOW_REMOVED:
3483 case OFPUTIL_NXT_PACKET_IN:
3484 case OFPUTIL_NXST_FLOW_REPLY:
3485 case OFPUTIL_NXST_AGGREGATE_REPLY:
3487 return (oh->type == OFPT10_STATS_REQUEST ||
3488 oh->type == OFPT10_STATS_REPLY
3489 ? OFPERR_OFPBRC_BAD_STAT
3490 : OFPERR_OFPBRC_BAD_TYPE);
3495 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3497 int error = handle_openflow__(ofconn, ofp_msg);
3498 if (error && error != OFPROTO_POSTPONE) {
3499 ofconn_send_error(ofconn, ofp_msg->data, error);
3501 COVERAGE_INC(ofproto_recv_openflow);
3502 return error != OFPROTO_POSTPONE;
3505 /* Asynchronous operations. */
3507 /* Creates and returns a new ofopgroup that is not associated with any
3508 * OpenFlow connection.
3510 * The caller should add operations to the returned group with
3511 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3512 static struct ofopgroup *
3513 ofopgroup_create_unattached(struct ofproto *ofproto)
3515 struct ofopgroup *group = xzalloc(sizeof *group);
3516 group->ofproto = ofproto;
3517 list_init(&group->ofproto_node);
3518 list_init(&group->ops);
3519 list_init(&group->ofconn_node);
3523 /* Creates and returns a new ofopgroup for 'ofproto'.
3525 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3526 * connection. The 'request' and 'buffer_id' arguments are ignored.
3528 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3529 * If the ofopgroup eventually fails, then the error reply will include
3530 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3531 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3533 * The caller should add operations to the returned group with
3534 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3535 static struct ofopgroup *
3536 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3537 const struct ofp_header *request, uint32_t buffer_id)
3539 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3541 size_t request_len = ntohs(request->length);
3543 assert(ofconn_get_ofproto(ofconn) == ofproto);
3545 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3546 group->ofconn = ofconn;
3547 group->request = xmemdup(request, MIN(request_len, 64));
3548 group->buffer_id = buffer_id;
3553 /* Submits 'group' for processing.
3555 * If 'group' contains no operations (e.g. none were ever added, or all of the
3556 * ones that were added completed synchronously), then it is destroyed
3557 * immediately. Otherwise it is added to the ofproto's list of pending
3560 ofopgroup_submit(struct ofopgroup *group)
3562 if (list_is_empty(&group->ops)) {
3563 ofopgroup_destroy(group);
3565 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3566 group->ofproto->n_pending++;
3571 ofopgroup_destroy(struct ofopgroup *group)
3573 assert(list_is_empty(&group->ops));
3574 if (!list_is_empty(&group->ofproto_node)) {
3575 assert(group->ofproto->n_pending > 0);
3576 group->ofproto->n_pending--;
3577 list_remove(&group->ofproto_node);
3579 if (!list_is_empty(&group->ofconn_node)) {
3580 list_remove(&group->ofconn_node);
3582 ofconn_send_error(group->ofconn, group->request, group->error);
3584 connmgr_retry(group->ofproto->connmgr);
3586 free(group->request);
3590 /* Initiates a new operation on 'rule', of the specified 'type', within
3591 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3593 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3594 enum ofoperation_type type)
3596 struct ofproto *ofproto = group->ofproto;
3597 struct ofoperation *op;
3599 assert(!rule->pending);
3601 op = rule->pending = xzalloc(sizeof *op);
3603 list_push_back(&group->ops, &op->group_node);
3606 op->flow_cookie = rule->flow_cookie;
3608 if (type == OFOPERATION_DELETE) {
3609 hmap_insert(&ofproto->deletions, &op->hmap_node,
3610 cls_rule_hash(&rule->cr, rule->table_id));
3615 ofoperation_destroy(struct ofoperation *op)
3617 struct ofopgroup *group = op->group;
3620 op->rule->pending = NULL;
3622 if (op->type == OFOPERATION_DELETE) {
3623 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3625 list_remove(&op->group_node);
3629 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3630 ofopgroup_destroy(group);
3634 /* Indicates that 'op' completed with status 'error', which is either 0 to
3635 * indicate success or an OpenFlow error code on failure.
3637 * If 'error' is 0, indicating success, the operation will be committed
3638 * permanently to the flow table. There is one interesting subcase:
3640 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3641 * the flow table (the "victim" rule) by a new one, then the caller must
3642 * have uninitialized any derived state in the victim rule, as in step 5 in
3643 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3644 * performs steps 6 and 7 for the victim rule, most notably by calling its
3645 * ->rule_dealloc() function.
3647 * If 'error' is nonzero, then generally the operation will be rolled back:
3649 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3650 * restores the original rule. The caller must have uninitialized any
3651 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3652 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3653 * and 7 for the new rule, calling its ->rule_dealloc() function.
3655 * - If 'op' is a "modify flow" operation, ofproto restores the original
3658 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3659 * allowed to fail. It must always succeed.
3661 * Please see the large comment in ofproto/ofproto-provider.h titled
3662 * "Asynchronous Operation Support" for more information. */
3664 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3666 struct ofopgroup *group = op->group;
3667 struct rule *rule = op->rule;
3668 struct ofproto *ofproto = rule->ofproto;
3670 assert(rule->pending == op);
3674 && op->type != OFOPERATION_DELETE
3676 && group->buffer_id != UINT32_MAX
3677 && list_is_singleton(&op->group_node)) {
3678 struct ofpbuf *packet;
3681 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3685 error = rule_execute(rule, in_port, packet);
3688 if (!group->error) {
3689 group->error = error;
3693 case OFOPERATION_ADD:
3695 ofproto_rule_destroy__(op->victim);
3696 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3697 == htons(VLAN_VID_MASK)) {
3698 if (ofproto->vlan_bitmap) {
3699 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3701 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3702 bitmap_set1(ofproto->vlan_bitmap, vid);
3703 ofproto->vlans_changed = true;
3706 ofproto->vlans_changed = true;
3710 oftable_substitute_rule(rule, op->victim);
3711 ofproto_rule_destroy__(rule);
3716 case OFOPERATION_DELETE:
3718 ofproto_rule_destroy__(rule);
3722 case OFOPERATION_MODIFY:
3724 rule->modified = time_msec();
3726 free(rule->ofpacts);
3727 rule->ofpacts = op->ofpacts;
3728 rule->ofpacts_len = op->ofpacts_len;
3736 ofoperation_destroy(op);
3740 ofoperation_get_victim(struct ofoperation *op)
3742 assert(op->type == OFOPERATION_ADD);
3747 pick_datapath_id(const struct ofproto *ofproto)
3749 const struct ofport *port;
3751 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3753 uint8_t ea[ETH_ADDR_LEN];
3756 error = netdev_get_etheraddr(port->netdev, ea);
3758 return eth_addr_to_uint64(ea);
3760 VLOG_WARN("could not get MAC address for %s (%s)",
3761 netdev_get_name(port->netdev), strerror(error));
3763 return ofproto->fallback_dpid;
3767 pick_fallback_dpid(void)
3769 uint8_t ea[ETH_ADDR_LEN];
3770 eth_addr_nicira_random(ea);
3771 return eth_addr_to_uint64(ea);
3774 /* Table overflow policy. */
3776 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3777 * is not configured to evict rules or if the table contains no evictable
3778 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3780 static struct rule *
3781 choose_rule_to_evict(struct oftable *table)
3783 struct eviction_group *evg;
3785 if (!table->eviction_fields) {
3789 /* In the common case, the outer and inner loops here will each be entered
3792 * - The inner loop normally "return"s in its first iteration. If the
3793 * eviction group has any evictable rules, then it always returns in
3796 * - The outer loop only iterates more than once if the largest eviction
3797 * group has no evictable rules.
3799 * - The outer loop can exit only if table's 'max_flows' is all filled up
3800 * by unevictable rules'. */
3801 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
3804 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
3805 if (rule->evictable) {
3814 /* Searches 'ofproto' for tables that have more flows than their configured
3815 * maximum and that have flow eviction enabled, and evicts as many flows as
3816 * necessary and currently feasible from them.
3818 * This triggers only when an OpenFlow table has N flows in it and then the
3819 * client configures a maximum number of flows less than N. */
3821 ofproto_evict(struct ofproto *ofproto)
3823 struct ofopgroup *group;
3824 struct oftable *table;
3826 group = ofopgroup_create_unattached(ofproto);
3827 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
3828 while (classifier_count(&table->cls) > table->max_flows
3829 && table->eviction_fields) {
3832 rule = choose_rule_to_evict(table);
3833 if (!rule || rule->pending) {
3837 ofoperation_create(group, rule, OFOPERATION_DELETE);
3838 oftable_remove_rule(rule);
3839 ofproto->ofproto_class->rule_destruct(rule);
3842 ofopgroup_submit(group);
3845 /* Eviction groups. */
3847 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3848 * rules. The priority contains low-order random bits to ensure that eviction
3849 * groups with the same number of rules are prioritized randomly. */
3851 eviction_group_priority(size_t n_rules)
3853 uint16_t size = MIN(UINT16_MAX, n_rules);
3854 return (size << 16) | random_uint16();
3857 /* Updates 'evg', an eviction_group within 'table', following a change that
3858 * adds or removes rules in 'evg'. */
3860 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
3862 heap_change(&table->eviction_groups_by_size, &evg->size_node,
3863 eviction_group_priority(heap_count(&evg->rules)));
3866 /* Destroys 'evg', an eviction_group within 'table':
3868 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3869 * rules themselves, just removes them from the eviction group.)
3871 * - Removes 'evg' from 'table'.
3875 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
3877 while (!heap_is_empty(&evg->rules)) {
3880 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
3881 rule->eviction_group = NULL;
3883 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
3884 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
3885 heap_destroy(&evg->rules);
3889 /* Removes 'rule' from its eviction group, if any. */
3891 eviction_group_remove_rule(struct rule *rule)
3893 if (rule->eviction_group) {
3894 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3895 struct eviction_group *evg = rule->eviction_group;
3897 rule->eviction_group = NULL;
3898 heap_remove(&evg->rules, &rule->evg_node);
3899 if (heap_is_empty(&evg->rules)) {
3900 eviction_group_destroy(table, evg);
3902 eviction_group_resized(table, evg);
3907 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3908 * returns the hash value. */
3910 eviction_group_hash_rule(struct rule *rule)
3912 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3913 const struct mf_subfield *sf;
3916 hash = table->eviction_group_id_basis;
3917 for (sf = table->eviction_fields;
3918 sf < &table->eviction_fields[table->n_eviction_fields];
3921 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
3922 union mf_value value;
3924 mf_get_value(sf->field, &rule->cr.flow, &value);
3926 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
3928 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
3929 unsigned int start = sf->ofs + sf->n_bits;
3930 bitwise_zero(&value, sf->field->n_bytes, start,
3931 sf->field->n_bytes * 8 - start);
3933 hash = hash_bytes(&value, sf->field->n_bytes, hash);
3935 hash = hash_int(hash, 0);
3942 /* Returns an eviction group within 'table' with the given 'id', creating one
3944 static struct eviction_group *
3945 eviction_group_find(struct oftable *table, uint32_t id)
3947 struct eviction_group *evg;
3949 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
3953 evg = xmalloc(sizeof *evg);
3954 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
3955 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
3956 eviction_group_priority(0));
3957 heap_init(&evg->rules);
3962 /* Returns an eviction priority for 'rule'. The return value should be
3963 * interpreted so that higher priorities make a rule more attractive candidates
3966 rule_eviction_priority(struct rule *rule)
3968 long long int hard_expiration;
3969 long long int idle_expiration;
3970 long long int expiration;
3971 uint32_t expiration_offset;
3973 /* Calculate time of expiration. */
3974 hard_expiration = (rule->hard_timeout
3975 ? rule->modified + rule->hard_timeout * 1000
3977 idle_expiration = (rule->idle_timeout
3978 ? rule->used + rule->idle_timeout * 1000
3980 expiration = MIN(hard_expiration, idle_expiration);
3981 if (expiration == LLONG_MAX) {
3985 /* Calculate the time of expiration as a number of (approximate) seconds
3986 * after program startup.
3988 * This should work OK for program runs that last UINT32_MAX seconds or
3989 * less. Therefore, please restart OVS at least once every 136 years. */
3990 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
3992 /* Invert the expiration offset because we're using a max-heap. */
3993 return UINT32_MAX - expiration_offset;
3996 /* Adds 'rule' to an appropriate eviction group for its oftable's
3997 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
3998 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4001 * The caller must ensure that 'rule' is not already in an eviction group. */
4003 eviction_group_add_rule(struct rule *rule)
4005 struct ofproto *ofproto = rule->ofproto;
4006 struct oftable *table = &ofproto->tables[rule->table_id];
4008 if (table->eviction_fields
4009 && (rule->hard_timeout || rule->idle_timeout)) {
4010 struct eviction_group *evg;
4012 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4014 rule->eviction_group = evg;
4015 heap_insert(&evg->rules, &rule->evg_node,
4016 rule_eviction_priority(rule));
4017 eviction_group_resized(table, evg);
4023 /* Initializes 'table'. */
4025 oftable_init(struct oftable *table)
4027 memset(table, 0, sizeof *table);
4028 classifier_init(&table->cls);
4029 table->max_flows = UINT_MAX;
4032 /* Destroys 'table', including its classifier and eviction groups.
4034 * The caller is responsible for freeing 'table' itself. */
4036 oftable_destroy(struct oftable *table)
4038 assert(classifier_is_empty(&table->cls));
4039 oftable_disable_eviction(table);
4040 classifier_destroy(&table->cls);
4044 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4045 * string, then 'table' will use its default name.
4047 * This only affects the name exposed for a table exposed through the OpenFlow
4048 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4050 oftable_set_name(struct oftable *table, const char *name)
4052 if (name && name[0]) {
4053 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4054 if (!table->name || strncmp(name, table->name, len)) {
4056 table->name = xmemdup0(name, len);
4064 /* oftables support a choice of two policies when adding a rule would cause the
4065 * number of flows in the table to exceed the configured maximum number: either
4066 * they can refuse to add the new flow or they can evict some existing flow.
4067 * This function configures the former policy on 'table'. */
4069 oftable_disable_eviction(struct oftable *table)
4071 if (table->eviction_fields) {
4072 struct eviction_group *evg, *next;
4074 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4075 &table->eviction_groups_by_id) {
4076 eviction_group_destroy(table, evg);
4078 hmap_destroy(&table->eviction_groups_by_id);
4079 heap_destroy(&table->eviction_groups_by_size);
4081 free(table->eviction_fields);
4082 table->eviction_fields = NULL;
4083 table->n_eviction_fields = 0;
4087 /* oftables support a choice of two policies when adding a rule would cause the
4088 * number of flows in the table to exceed the configured maximum number: either
4089 * they can refuse to add the new flow or they can evict some existing flow.
4090 * This function configures the latter policy on 'table', with fairness based
4091 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4092 * 'n_fields' as 0 disables fairness.) */
4094 oftable_enable_eviction(struct oftable *table,
4095 const struct mf_subfield *fields, size_t n_fields)
4097 struct cls_cursor cursor;
4100 if (table->eviction_fields
4101 && n_fields == table->n_eviction_fields
4103 || !memcmp(fields, table->eviction_fields,
4104 n_fields * sizeof *fields))) {
4109 oftable_disable_eviction(table);
4111 table->n_eviction_fields = n_fields;
4112 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4114 table->eviction_group_id_basis = random_uint32();
4115 hmap_init(&table->eviction_groups_by_id);
4116 heap_init(&table->eviction_groups_by_size);
4118 cls_cursor_init(&cursor, &table->cls, NULL);
4119 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4120 eviction_group_add_rule(rule);
4124 /* Removes 'rule' from the oftable that contains it. */
4126 oftable_remove_rule(struct rule *rule)
4128 struct ofproto *ofproto = rule->ofproto;
4129 struct oftable *table = &ofproto->tables[rule->table_id];
4131 classifier_remove(&table->cls, &rule->cr);
4132 eviction_group_remove_rule(rule);
4135 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4136 * oftable that has an identical cls_rule. Returns the rule that was removed,
4137 * if any, and otherwise NULL. */
4138 static struct rule *
4139 oftable_replace_rule(struct rule *rule)
4141 struct ofproto *ofproto = rule->ofproto;
4142 struct oftable *table = &ofproto->tables[rule->table_id];
4143 struct rule *victim;
4145 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4147 eviction_group_remove_rule(victim);
4149 eviction_group_add_rule(rule);
4153 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4155 oftable_substitute_rule(struct rule *old, struct rule *new)
4158 oftable_replace_rule(new);
4160 oftable_remove_rule(old);
4164 /* unixctl commands. */
4167 ofproto_lookup(const char *name)
4169 struct ofproto *ofproto;
4171 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4173 if (!strcmp(ofproto->name, name)) {
4181 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4182 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4184 struct ofproto *ofproto;
4188 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4189 ds_put_format(&results, "%s\n", ofproto->name);
4191 unixctl_command_reply(conn, ds_cstr(&results));
4192 ds_destroy(&results);
4196 ofproto_unixctl_init(void)
4198 static bool registered;
4204 unixctl_command_register("ofproto/list", "", 0, 0,
4205 ofproto_unixctl_list, NULL);
4208 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4210 * This is deprecated. It is only for compatibility with broken device drivers
4211 * in old versions of Linux that do not properly support VLANs when VLAN
4212 * devices are not used. When broken device drivers are no longer in
4213 * widespread use, we will delete these interfaces. */
4215 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4216 * (exactly) by an OpenFlow rule in 'ofproto'. */
4218 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4220 const struct oftable *oftable;
4222 free(ofproto->vlan_bitmap);
4223 ofproto->vlan_bitmap = bitmap_allocate(4096);
4224 ofproto->vlans_changed = false;
4226 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4227 const struct cls_table *table;
4229 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4230 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4231 == htons(VLAN_VID_MASK)) {
4232 const struct cls_rule *rule;
4234 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4235 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4236 bitmap_set1(vlan_bitmap, vid);
4237 bitmap_set1(ofproto->vlan_bitmap, vid);
4244 /* Returns true if new VLANs have come into use by the flow table since the
4245 * last call to ofproto_get_vlan_usage().
4247 * We don't track when old VLANs stop being used. */
4249 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4251 return ofproto->vlans_changed;
4254 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4255 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4256 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4257 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4258 * then the VLAN device is un-enslaved. */
4260 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4261 uint16_t realdev_ofp_port, int vid)
4263 struct ofport *ofport;
4266 assert(vlandev_ofp_port != realdev_ofp_port);
4268 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4270 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4271 ofproto->name, vlandev_ofp_port);
4275 if (!ofproto->ofproto_class->set_realdev) {
4276 if (!vlandev_ofp_port) {
4279 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4283 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4285 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4286 ofproto->name, vlandev_ofp_port,
4287 netdev_get_name(ofport->netdev), strerror(error));