2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
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.
24 #include "byte-order.h"
25 #include "classifier.h"
28 #include "dynamic-string.h"
33 #include "ofp-print.h"
36 #include "ofproto-provider.h"
37 #include "openflow/nicira-ext.h"
38 #include "openflow/openflow.h"
42 #include "poll-loop.h"
46 #include "unaligned.h"
50 VLOG_DEFINE_THIS_MODULE(ofproto);
52 COVERAGE_DEFINE(ofproto_error);
53 COVERAGE_DEFINE(ofproto_flush);
54 COVERAGE_DEFINE(ofproto_no_packet_in);
55 COVERAGE_DEFINE(ofproto_packet_out);
56 COVERAGE_DEFINE(ofproto_queue_req);
57 COVERAGE_DEFINE(ofproto_recv_openflow);
58 COVERAGE_DEFINE(ofproto_reinit_ports);
59 COVERAGE_DEFINE(ofproto_uninstallable);
60 COVERAGE_DEFINE(ofproto_update_port);
63 S_OPENFLOW, /* Processing OpenFlow commands. */
64 S_FLUSH, /* Deleting all flow table rules. */
67 enum ofoperation_type {
73 /* A single OpenFlow request can execute any number of operations. The
74 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
75 * ofconn to which an error reply should be sent if necessary.
77 * ofproto initiates some operations internally. These operations are still
78 * assigned to groups but will not have an associated ofconn. */
80 struct ofproto *ofproto; /* Owning ofproto. */
81 struct list ofproto_node; /* In ofproto's "pending" list. */
82 struct list ops; /* List of "struct ofoperation"s. */
84 /* Data needed to send OpenFlow reply on failure or to send a buffered
87 * If list_is_empty(ofconn_node) then this ofopgroup never had an
88 * associated ofconn or its ofconn's connection dropped after it initiated
89 * the operation. In the latter case 'ofconn' is a wild pointer that
90 * refers to freed memory, so the 'ofconn' member must be used only if
91 * !list_is_empty(ofconn_node).
93 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
94 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
95 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
96 uint32_t buffer_id; /* Buffer id from original request. */
97 int error; /* 0 if no error yet, otherwise error code. */
100 static struct ofopgroup *ofopgroup_create(struct ofproto *);
101 static struct ofopgroup *ofopgroup_create_for_ofconn(struct ofconn *,
102 const struct ofp_header *,
104 static void ofopgroup_submit(struct ofopgroup *);
105 static void ofopgroup_destroy(struct ofopgroup *);
107 /* A single flow table operation. */
109 struct ofopgroup *group; /* Owning group. */
110 struct list group_node; /* In ofopgroup's "ops" list. */
111 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
112 struct rule *rule; /* Rule being operated upon. */
113 enum ofoperation_type type; /* Type of operation. */
114 int status; /* -1 if pending, otherwise 0 or error code. */
115 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
116 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
117 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
118 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
121 static void ofoperation_create(struct ofopgroup *, struct rule *,
122 enum ofoperation_type);
123 static void ofoperation_destroy(struct ofoperation *);
125 static void ofport_destroy__(struct ofport *);
126 static void ofport_destroy(struct ofport *);
128 static uint64_t pick_datapath_id(const struct ofproto *);
129 static uint64_t pick_fallback_dpid(void);
131 static void ofproto_destroy__(struct ofproto *);
133 static void ofproto_rule_destroy__(struct rule *);
134 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
136 static void ofopgroup_destroy(struct ofopgroup *);
138 static int add_flow(struct ofproto *, struct ofconn *, struct flow_mod *,
139 const struct ofp_header *);
141 /* This return value tells handle_openflow() that processing of the current
142 * OpenFlow message must be postponed until some ongoing operations have
145 * This particular value is a good choice because it is negative (so it won't
146 * collide with any errno value or any value returned by ofp_mkerr()) and large
147 * (so it won't accidentally collide with EOF or a negative errno value). */
148 enum { OFPROTO_POSTPONE = -100000 };
150 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
152 static void update_port(struct ofproto *, const char *devname);
153 static int init_ports(struct ofproto *);
154 static void reinit_ports(struct ofproto *);
156 static void ofproto_unixctl_init(void);
158 /* All registered ofproto classes, in probe order. */
159 static const struct ofproto_class **ofproto_classes;
160 static size_t n_ofproto_classes;
161 static size_t allocated_ofproto_classes;
163 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
164 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
166 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
169 ofproto_initialize(void)
175 ofproto_class_register(&ofproto_dpif_class);
179 /* 'type' should be a normalized datapath type, as returned by
180 * ofproto_normalize_type(). Returns the corresponding ofproto_class
181 * structure, or a null pointer if there is none registered for 'type'. */
182 static const struct ofproto_class *
183 ofproto_class_find__(const char *type)
187 ofproto_initialize();
188 for (i = 0; i < n_ofproto_classes; i++) {
189 const struct ofproto_class *class = ofproto_classes[i];
194 class->enumerate_types(&types);
195 found = sset_contains(&types, type);
196 sset_destroy(&types);
202 VLOG_WARN("unknown datapath type %s", type);
206 /* Registers a new ofproto class. After successful registration, new ofprotos
207 * of that type can be created using ofproto_create(). */
209 ofproto_class_register(const struct ofproto_class *new_class)
213 for (i = 0; i < n_ofproto_classes; i++) {
214 if (ofproto_classes[i] == new_class) {
219 if (n_ofproto_classes >= allocated_ofproto_classes) {
220 ofproto_classes = x2nrealloc(ofproto_classes,
221 &allocated_ofproto_classes,
222 sizeof *ofproto_classes);
224 ofproto_classes[n_ofproto_classes++] = new_class;
228 /* Unregisters a datapath provider. 'type' must have been previously
229 * registered and not currently be in use by any ofprotos. After
230 * unregistration new datapaths of that type cannot be opened using
231 * ofproto_create(). */
233 ofproto_class_unregister(const struct ofproto_class *class)
237 for (i = 0; i < n_ofproto_classes; i++) {
238 if (ofproto_classes[i] == class) {
239 for (i++; i < n_ofproto_classes; i++) {
240 ofproto_classes[i - 1] = ofproto_classes[i];
246 VLOG_WARN("attempted to unregister an ofproto class that is not "
251 /* Clears 'types' and enumerates all registered ofproto types into it. The
252 * caller must first initialize the sset. */
254 ofproto_enumerate_types(struct sset *types)
258 ofproto_initialize();
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->enumerate_types(types);
264 /* Returns the fully spelled out name for the given ofproto 'type'.
266 * Normalized type string can be compared with strcmp(). Unnormalized type
267 * string might be the same even if they have different spellings. */
269 ofproto_normalize_type(const char *type)
271 return type && type[0] ? type : "system";
274 /* Clears 'names' and enumerates the names of all known created ofprotos with
275 * the given 'type'. The caller must first initialize the sset. Returns 0 if
276 * successful, otherwise a positive errno value.
278 * Some kinds of datapaths might not be practically enumerable. This is not
279 * considered an error. */
281 ofproto_enumerate_names(const char *type, struct sset *names)
283 const struct ofproto_class *class = ofproto_class_find__(type);
284 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
288 ofproto_create(const char *datapath_name, const char *datapath_type,
289 struct ofproto **ofprotop)
291 const struct ofproto_class *class;
292 struct ofproto *ofproto;
297 ofproto_initialize();
298 ofproto_unixctl_init();
300 datapath_type = ofproto_normalize_type(datapath_type);
301 class = ofproto_class_find__(datapath_type);
303 VLOG_WARN("could not create datapath %s of unknown type %s",
304 datapath_name, datapath_type);
308 ofproto = class->alloc();
310 VLOG_ERR("failed to allocate datapath %s of type %s",
311 datapath_name, datapath_type);
316 memset(ofproto, 0, sizeof *ofproto);
317 ofproto->ofproto_class = class;
318 ofproto->name = xstrdup(datapath_name);
319 ofproto->type = xstrdup(datapath_type);
320 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
321 hash_string(ofproto->name, 0));
322 ofproto->datapath_id = 0;
323 ofproto_set_flow_eviction_threshold(ofproto,
324 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
325 ofproto->fallback_dpid = pick_fallback_dpid();
326 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
327 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
328 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
329 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
330 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
331 hmap_init(&ofproto->ports);
332 shash_init(&ofproto->port_by_name);
333 ofproto->tables = NULL;
334 ofproto->n_tables = 0;
335 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
336 ofproto->state = S_OPENFLOW;
337 list_init(&ofproto->pending);
338 hmap_init(&ofproto->deletions);
340 error = ofproto->ofproto_class->construct(ofproto);
342 VLOG_ERR("failed to open datapath %s: %s",
343 datapath_name, strerror(error));
344 ofproto_destroy__(ofproto);
347 assert(ofproto->n_tables > 0);
349 ofproto->datapath_id = pick_datapath_id(ofproto);
350 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
358 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
360 uint64_t old_dpid = p->datapath_id;
361 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
362 if (p->datapath_id != old_dpid) {
363 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
365 /* Force all active connections to reconnect, since there is no way to
366 * notify a controller that the datapath ID has changed. */
367 ofproto_reconnect_controllers(p);
372 ofproto_set_controllers(struct ofproto *p,
373 const struct ofproto_controller *controllers,
374 size_t n_controllers)
376 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
380 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
382 connmgr_set_fail_mode(p->connmgr, fail_mode);
385 /* Drops the connections between 'ofproto' and all of its controllers, forcing
386 * them to reconnect. */
388 ofproto_reconnect_controllers(struct ofproto *ofproto)
390 connmgr_reconnect(ofproto->connmgr);
393 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
394 * in-band control should guarantee access, in the same way that in-band
395 * control guarantees access to OpenFlow controllers. */
397 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
398 const struct sockaddr_in *extras, size_t n)
400 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
403 /* Sets the OpenFlow queue used by flows set up by in-band control on
404 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
405 * flows will use the default queue. */
407 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
409 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
412 /* Sets the number of flows at which eviction from the kernel flow table
415 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
417 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
418 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
420 ofproto->flow_eviction_threshold = threshold;
425 ofproto_set_desc(struct ofproto *p,
426 const char *mfr_desc, const char *hw_desc,
427 const char *sw_desc, const char *serial_desc,
430 struct ofp_desc_stats *ods;
433 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
434 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
435 sizeof ods->mfr_desc);
438 p->mfr_desc = xstrdup(mfr_desc);
441 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
442 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
443 sizeof ods->hw_desc);
446 p->hw_desc = xstrdup(hw_desc);
449 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
450 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
451 sizeof ods->sw_desc);
454 p->sw_desc = xstrdup(sw_desc);
457 if (strlen(serial_desc) >= sizeof ods->serial_num) {
458 VLOG_WARN("truncating serial_desc, must be less than %zu "
460 sizeof ods->serial_num);
462 free(p->serial_desc);
463 p->serial_desc = xstrdup(serial_desc);
466 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
467 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
468 sizeof ods->dp_desc);
471 p->dp_desc = xstrdup(dp_desc);
476 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
478 return connmgr_set_snoops(ofproto->connmgr, snoops);
482 ofproto_set_netflow(struct ofproto *ofproto,
483 const struct netflow_options *nf_options)
485 if (nf_options && sset_is_empty(&nf_options->collectors)) {
489 if (ofproto->ofproto_class->set_netflow) {
490 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
492 return nf_options ? EOPNOTSUPP : 0;
497 ofproto_set_sflow(struct ofproto *ofproto,
498 const struct ofproto_sflow_options *oso)
500 if (oso && sset_is_empty(&oso->targets)) {
504 if (ofproto->ofproto_class->set_sflow) {
505 return ofproto->ofproto_class->set_sflow(ofproto, oso);
507 return oso ? EOPNOTSUPP : 0;
511 /* Connectivity Fault Management configuration. */
513 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
515 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
517 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
518 if (ofport && ofproto->ofproto_class->set_cfm) {
519 ofproto->ofproto_class->set_cfm(ofport, NULL);
523 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
524 * basic configuration from the configuration members in 'cfm', and the remote
525 * maintenance point ID from remote_mpid. Ignores the statistics members of
528 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
530 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
531 const struct cfm_settings *s)
533 struct ofport *ofport;
536 ofport = ofproto_get_port(ofproto, ofp_port);
538 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
539 ofproto->name, ofp_port);
543 /* XXX: For configuration simplicity, we only support one remote_mpid
544 * outside of the CFM module. It's not clear if this is the correct long
545 * term solution or not. */
546 error = (ofproto->ofproto_class->set_cfm
547 ? ofproto->ofproto_class->set_cfm(ofport, s)
550 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
551 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
556 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
557 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
558 * 0 if LACP partner information is not current (generally indicating a
559 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
561 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
563 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
564 return (ofport && ofproto->ofproto_class->port_is_lacp_current
565 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
571 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
572 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
573 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
574 * configuration plus, if there is more than one slave, a bonding
577 * If 'aux' is already registered then this function updates its configuration
578 * to 's'. Otherwise, this function registers a new bundle.
580 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
583 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
584 const struct ofproto_bundle_settings *s)
586 return (ofproto->ofproto_class->bundle_set
587 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
591 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
592 * If no such bundle has been registered, this has no effect. */
594 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
596 return ofproto_bundle_register(ofproto, aux, NULL);
600 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
601 * If 'aux' is already registered then this function updates its configuration
602 * to 's'. Otherwise, this function registers a new mirror.
604 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
607 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
608 const struct ofproto_mirror_settings *s)
610 return (ofproto->ofproto_class->mirror_set
611 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
615 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
616 * If no mirror has been registered, this has no effect. */
618 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
620 return ofproto_mirror_register(ofproto, aux, NULL);
623 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
624 * which all packets are flooded, instead of using MAC learning. If
625 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
627 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
630 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
632 return (ofproto->ofproto_class->set_flood_vlans
633 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
637 /* Returns true if 'aux' is a registered bundle that is currently in use as the
638 * output for a mirror. */
640 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
642 return (ofproto->ofproto_class->is_mirror_output_bundle
643 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
648 ofproto_has_snoops(const struct ofproto *ofproto)
650 return connmgr_has_snoops(ofproto->connmgr);
654 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
656 connmgr_get_snoops(ofproto->connmgr, snoops);
660 ofproto_flush__(struct ofproto *ofproto)
662 struct classifier *table;
663 struct ofopgroup *group;
665 if (ofproto->ofproto_class->flush) {
666 ofproto->ofproto_class->flush(ofproto);
669 group = ofopgroup_create(ofproto);
670 for (table = ofproto->tables; table < &ofproto->tables[ofproto->n_tables];
672 struct rule *rule, *next_rule;
673 struct cls_cursor cursor;
675 cls_cursor_init(&cursor, table, NULL);
676 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
677 if (!rule->pending) {
678 ofoperation_create(group, rule, OFOPERATION_DELETE);
679 classifier_remove(table, &rule->cr);
680 ofproto->ofproto_class->rule_destruct(rule);
684 ofopgroup_submit(group);
688 ofproto_destroy__(struct ofproto *ofproto)
692 assert(list_is_empty(&ofproto->pending));
694 connmgr_destroy(ofproto->connmgr);
696 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
699 free(ofproto->mfr_desc);
700 free(ofproto->hw_desc);
701 free(ofproto->sw_desc);
702 free(ofproto->serial_desc);
703 free(ofproto->dp_desc);
704 hmap_destroy(&ofproto->ports);
705 shash_destroy(&ofproto->port_by_name);
707 for (i = 0; i < ofproto->n_tables; i++) {
708 assert(classifier_is_empty(&ofproto->tables[i]));
709 classifier_destroy(&ofproto->tables[i]);
711 free(ofproto->tables);
713 hmap_destroy(&ofproto->deletions);
715 ofproto->ofproto_class->dealloc(ofproto);
719 ofproto_destroy(struct ofproto *p)
721 struct ofport *ofport, *next_ofport;
728 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
729 ofport_destroy(ofport);
732 p->ofproto_class->destruct(p);
733 ofproto_destroy__(p);
736 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
737 * kernel datapath, for example, this destroys the datapath in the kernel, and
738 * with the netdev-based datapath, it tears down the data structures that
739 * represent the datapath.
741 * The datapath should not be currently open as an ofproto. */
743 ofproto_delete(const char *name, const char *type)
745 const struct ofproto_class *class = ofproto_class_find__(type);
746 return (!class ? EAFNOSUPPORT
747 : !class->del ? EACCES
748 : class->del(type, name));
752 process_port_change(struct ofproto *ofproto, int error, char *devname)
754 if (error == ENOBUFS) {
755 reinit_ports(ofproto);
757 update_port(ofproto, devname);
763 ofproto_run(struct ofproto *p)
765 struct ofport *ofport;
769 error = p->ofproto_class->run(p);
770 if (error == ENODEV) {
771 /* Someone destroyed the datapath behind our back. The caller
772 * better destroy us and give up, because we're just going to
773 * spin from here on out. */
774 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
775 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
780 if (p->ofproto_class->port_poll) {
781 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
782 process_port_change(p, error, devname);
786 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
787 unsigned int change_seq = netdev_change_seq(ofport->netdev);
788 if (ofport->change_seq != change_seq) {
789 ofport->change_seq = change_seq;
790 update_port(p, netdev_get_name(ofport->netdev));
797 connmgr_run(p->connmgr, handle_openflow);
801 connmgr_run(p->connmgr, NULL);
803 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
804 connmgr_flushed(p->connmgr);
805 p->state = S_OPENFLOW;
817 ofproto_wait(struct ofproto *p)
819 struct ofport *ofport;
821 p->ofproto_class->wait(p);
822 if (p->ofproto_class->port_poll_wait) {
823 p->ofproto_class->port_poll_wait(p);
826 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
827 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
828 poll_immediate_wake();
834 connmgr_wait(p->connmgr, true);
838 connmgr_wait(p->connmgr, false);
839 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
840 poll_immediate_wake();
847 ofproto_is_alive(const struct ofproto *p)
849 return connmgr_has_controllers(p->connmgr);
853 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
856 connmgr_get_controller_info(ofproto->connmgr, info);
860 ofproto_free_ofproto_controller_info(struct shash *info)
862 connmgr_free_controller_info(info);
865 /* Makes a deep copy of 'old' into 'port'. */
867 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
869 port->name = xstrdup(old->name);
870 port->type = xstrdup(old->type);
871 port->ofp_port = old->ofp_port;
874 /* Frees memory allocated to members of 'ofproto_port'.
876 * Do not call this function on an ofproto_port obtained from
877 * ofproto_port_dump_next(): that function retains ownership of the data in the
880 ofproto_port_destroy(struct ofproto_port *ofproto_port)
882 free(ofproto_port->name);
883 free(ofproto_port->type);
886 /* Initializes 'dump' to begin dumping the ports in an ofproto.
888 * This function provides no status indication. An error status for the entire
889 * dump operation is provided when it is completed by calling
890 * ofproto_port_dump_done().
893 ofproto_port_dump_start(struct ofproto_port_dump *dump,
894 const struct ofproto *ofproto)
896 dump->ofproto = ofproto;
897 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
901 /* Attempts to retrieve another port from 'dump', which must have been created
902 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
903 * 'port' and returns true. On failure, returns false.
905 * Failure might indicate an actual error or merely that the last port has been
906 * dumped. An error status for the entire dump operation is provided when it
907 * is completed by calling ofproto_port_dump_done().
909 * The ofproto owns the data stored in 'port'. It will remain valid until at
910 * least the next time 'dump' is passed to ofproto_port_dump_next() or
911 * ofproto_port_dump_done(). */
913 ofproto_port_dump_next(struct ofproto_port_dump *dump,
914 struct ofproto_port *port)
916 const struct ofproto *ofproto = dump->ofproto;
922 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
925 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
931 /* Completes port table dump operation 'dump', which must have been created
932 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
933 * error-free, otherwise a positive errno value describing the problem. */
935 ofproto_port_dump_done(struct ofproto_port_dump *dump)
937 const struct ofproto *ofproto = dump->ofproto;
939 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
942 return dump->error == EOF ? 0 : dump->error;
945 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
946 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
947 * is non-null). On failure, returns a positive errno value and sets
948 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
950 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
956 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
958 update_port(ofproto, netdev_get_name(netdev));
961 *ofp_portp = error ? OFPP_NONE : ofp_port;
966 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
967 * initializes '*port' appropriately; on failure, returns a positive errno
970 * The caller owns the data in 'ofproto_port' and must free it with
971 * ofproto_port_destroy() when it is no longer needed. */
973 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
974 struct ofproto_port *port)
978 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
980 memset(port, 0, sizeof *port);
985 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
986 * Returns 0 if successful, otherwise a positive errno. */
988 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
990 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
991 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
994 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
995 if (!error && ofport) {
996 /* 'name' is the netdev's name and update_port() is going to close the
997 * netdev. Just in case update_port() refers to 'name' after it
998 * destroys 'ofport', make a copy of it around the update_port()
1000 char *devname = xstrdup(name);
1001 update_port(ofproto, devname);
1007 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1008 * performs the 'n_actions' actions in 'actions'. The new flow will not
1011 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1012 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1013 * controllers; otherwise, it will be hidden.
1015 * The caller retains ownership of 'cls_rule' and 'actions'.
1017 * This is a helper function for in-band control and fail-open. */
1019 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1020 const union ofp_action *actions, size_t n_actions)
1022 const struct rule *rule;
1024 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1025 &ofproto->tables[0], cls_rule));
1026 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1027 actions, n_actions)) {
1030 memset(&fm, 0, sizeof fm);
1032 fm.buffer_id = UINT32_MAX;
1033 fm.actions = (union ofp_action *) actions;
1034 fm.n_actions = n_actions;
1035 add_flow(ofproto, NULL, &fm, NULL);
1039 /* Searches for a rule with matching criteria exactly equal to 'target' in
1040 * ofproto's table 0 and, if it finds one, deletes it.
1042 * This is a helper function for in-band control and fail-open. */
1044 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1048 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1049 &ofproto->tables[0], target));
1051 /* No such rule -> success. */
1053 } else if (rule->pending) {
1054 /* An operation on the rule is already pending -> failure.
1055 * Caller must retry later if it's important. */
1058 /* Initiate deletion -> success. */
1059 struct ofopgroup *group = ofopgroup_create(ofproto);
1060 ofoperation_create(group, rule, OFOPERATION_DELETE);
1061 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1062 rule->ofproto->ofproto_class->rule_destruct(rule);
1063 ofopgroup_submit(group);
1069 /* Starts the process of deleting all of the flows from all of ofproto's flow
1070 * tables and then reintroducing the flows required by in-band control and
1071 * fail-open. The process will complete in a later call to ofproto_run(). */
1073 ofproto_flush_flows(struct ofproto *ofproto)
1075 COVERAGE_INC(ofproto_flush);
1076 ofproto->state = S_FLUSH;
1080 reinit_ports(struct ofproto *p)
1082 struct ofproto_port_dump dump;
1083 struct sset devnames;
1084 struct ofport *ofport;
1085 struct ofproto_port ofproto_port;
1086 const char *devname;
1088 COVERAGE_INC(ofproto_reinit_ports);
1090 sset_init(&devnames);
1091 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1092 sset_add(&devnames, netdev_get_name(ofport->netdev));
1094 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1095 sset_add(&devnames, ofproto_port.name);
1098 SSET_FOR_EACH (devname, &devnames) {
1099 update_port(p, devname);
1101 sset_destroy(&devnames);
1104 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1105 * netdev cannot be opened. On success, also fills in 'opp'. */
1106 static struct netdev *
1107 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1109 uint32_t curr, advertised, supported, peer;
1110 struct netdev_options netdev_options;
1111 enum netdev_flags flags;
1112 struct netdev *netdev;
1115 memset(&netdev_options, 0, sizeof netdev_options);
1116 netdev_options.name = ofproto_port->name;
1117 netdev_options.type = ofproto_port->type;
1118 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1120 error = netdev_open(&netdev_options, &netdev);
1122 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1123 "cannot be opened (%s)",
1124 ofproto_port->name, ofproto_port->ofp_port,
1125 ofproto_port->name, strerror(error));
1129 netdev_get_flags(netdev, &flags);
1130 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1132 opp->port_no = htons(ofproto_port->ofp_port);
1133 netdev_get_etheraddr(netdev, opp->hw_addr);
1134 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1135 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1136 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1137 opp->curr = htonl(curr);
1138 opp->advertised = htonl(advertised);
1139 opp->supported = htonl(supported);
1140 opp->peer = htonl(peer);
1145 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1146 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1149 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1151 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1152 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1153 && a->state == b->state
1154 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1155 && a->curr == b->curr
1156 && a->advertised == b->advertised
1157 && a->supported == b->supported
1158 && a->peer == b->peer);
1161 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1162 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1163 * one with the same name or port number). */
1165 ofport_install(struct ofproto *p,
1166 struct netdev *netdev, const struct ofp_phy_port *opp)
1168 const char *netdev_name = netdev_get_name(netdev);
1169 struct ofport *ofport;
1172 /* Create ofport. */
1173 ofport = p->ofproto_class->port_alloc();
1178 ofport->ofproto = p;
1179 ofport->netdev = netdev;
1180 ofport->change_seq = netdev_change_seq(netdev);
1182 ofport->ofp_port = ntohs(opp->port_no);
1184 /* Add port to 'p'. */
1185 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1186 shash_add(&p->port_by_name, netdev_name, ofport);
1188 /* Let the ofproto_class initialize its private data. */
1189 error = p->ofproto_class->port_construct(ofport);
1193 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1197 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1198 p->name, netdev_name, strerror(error));
1200 ofport_destroy__(ofport);
1202 netdev_close(netdev);
1206 /* Removes 'ofport' from 'p' and destroys it. */
1208 ofport_remove(struct ofport *ofport)
1210 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1212 ofport_destroy(ofport);
1215 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1218 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1220 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1222 ofport_remove(port);
1226 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1228 * Does not handle a name or port number change. The caller must implement
1229 * such a change as a delete followed by an add. */
1231 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1233 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1234 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1235 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1236 port->opp.state = opp->state;
1237 port->opp.curr = opp->curr;
1238 port->opp.advertised = opp->advertised;
1239 port->opp.supported = opp->supported;
1240 port->opp.peer = opp->peer;
1242 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1246 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1248 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1250 if (port->ofproto->ofproto_class->set_cfm) {
1251 port->ofproto->ofproto_class->set_cfm(port, NULL);
1253 if (port->ofproto->ofproto_class->bundle_remove) {
1254 port->ofproto->ofproto_class->bundle_remove(port);
1260 ofport_destroy__(struct ofport *port)
1262 struct ofproto *ofproto = port->ofproto;
1263 const char *name = netdev_get_name(port->netdev);
1265 hmap_remove(&ofproto->ports, &port->hmap_node);
1266 shash_delete(&ofproto->port_by_name,
1267 shash_find(&ofproto->port_by_name, name));
1269 netdev_close(port->netdev);
1270 ofproto->ofproto_class->port_dealloc(port);
1274 ofport_destroy(struct ofport *port)
1277 port->ofproto->ofproto_class->port_destruct(port);
1278 ofport_destroy__(port);
1283 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1285 struct ofport *port;
1287 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1288 hash_int(ofp_port, 0), &ofproto->ports) {
1289 if (port->ofp_port == ofp_port) {
1297 update_port(struct ofproto *ofproto, const char *name)
1299 struct ofproto_port ofproto_port;
1300 struct ofp_phy_port opp;
1301 struct netdev *netdev;
1302 struct ofport *port;
1304 COVERAGE_INC(ofproto_update_port);
1306 /* Fetch 'name''s location and properties from the datapath. */
1307 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1308 ? ofport_open(&ofproto_port, &opp)
1311 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1312 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1313 struct netdev *old_netdev = port->netdev;
1315 /* 'name' hasn't changed location. Any properties changed? */
1316 if (!ofport_equal(&port->opp, &opp)) {
1317 ofport_modified(port, &opp);
1320 /* Install the newly opened netdev in case it has changed.
1321 * Don't close the old netdev yet in case port_modified has to
1322 * remove a retained reference to it.*/
1323 port->netdev = netdev;
1324 port->change_seq = netdev_change_seq(netdev);
1326 if (port->ofproto->ofproto_class->port_modified) {
1327 port->ofproto->ofproto_class->port_modified(port);
1330 netdev_close(old_netdev);
1332 /* If 'port' is nonnull then its name differs from 'name' and thus
1333 * we should delete it. If we think there's a port named 'name'
1334 * then its port number must be wrong now so delete it too. */
1336 ofport_remove(port);
1338 ofport_remove_with_name(ofproto, name);
1339 ofport_install(ofproto, netdev, &opp);
1342 /* Any port named 'name' is gone now. */
1343 ofport_remove_with_name(ofproto, name);
1345 ofproto_port_destroy(&ofproto_port);
1349 init_ports(struct ofproto *p)
1351 struct ofproto_port_dump dump;
1352 struct ofproto_port ofproto_port;
1354 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1355 uint16_t ofp_port = ofproto_port.ofp_port;
1356 if (ofproto_get_port(p, ofp_port)) {
1357 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1359 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1360 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1363 struct ofp_phy_port opp;
1364 struct netdev *netdev;
1366 netdev = ofport_open(&ofproto_port, &opp);
1368 ofport_install(p, netdev, &opp);
1377 ofproto_rule_destroy__(struct rule *rule)
1379 free(rule->actions);
1380 rule->ofproto->ofproto_class->rule_dealloc(rule);
1383 /* This function allows an ofproto implementation to destroy any rules that
1384 * remain when its ->destruct() function is called. The caller must have
1385 * already uninitialized any derived members of 'rule' (step 5 described in the
1386 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1387 * This function implements steps 6 and 7.
1389 * This function should only be called from an ofproto implementation's
1390 * ->destruct() function. It is not suitable elsewhere. */
1392 ofproto_rule_destroy(struct rule *rule)
1394 assert(!rule->pending);
1395 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1396 ofproto_rule_destroy__(rule);
1399 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1400 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1403 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1405 const union ofp_action *oa;
1408 if (out_port == OFPP_NONE) {
1411 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1412 if (action_outputs_to_port(oa, htons(out_port))) {
1419 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1420 * statistics appropriately. 'packet' must have at least sizeof(struct
1421 * ofp_packet_in) bytes of headroom.
1423 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1424 * with statistics for 'packet' either way.
1426 * Takes ownership of 'packet'. */
1428 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1432 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1434 flow_extract(packet, 0, in_port, &flow);
1435 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1438 /* Returns true if 'rule' should be hidden from the controller.
1440 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1441 * (e.g. by in-band control) and are intentionally hidden from the
1444 rule_is_hidden(const struct rule *rule)
1446 return rule->cr.priority > UINT16_MAX;
1450 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1452 ofconn_send_reply(ofconn, make_echo_reply(oh));
1457 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1459 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1460 struct ofp_switch_features *osf;
1462 struct ofport *port;
1466 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1467 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1469 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1470 osf->datapath_id = htonll(ofproto->datapath_id);
1471 osf->n_buffers = htonl(pktbuf_capacity());
1472 osf->n_tables = ofproto->n_tables;
1473 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1476 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1478 osf->actions = htonl(actions);
1480 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1481 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1484 ofconn_send_reply(ofconn, buf);
1489 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1491 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1493 struct ofp_switch_config *osc;
1497 /* Figure out flags. */
1498 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1499 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1502 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1503 osc->flags = htons(flags);
1504 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1505 ofconn_send_reply(ofconn, buf);
1511 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1513 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1514 uint16_t flags = ntohs(osc->flags);
1516 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1517 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1518 switch (flags & OFPC_FRAG_MASK) {
1519 case OFPC_FRAG_NORMAL:
1520 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1522 case OFPC_FRAG_DROP:
1523 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1526 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1532 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1537 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1538 * error message code (composed with ofp_mkerr()) for the caller to propagate
1539 * upward. Otherwise, returns 0.
1541 * The log message mentions 'msg_type'. */
1543 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1545 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1546 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1547 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1548 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1551 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1558 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1560 struct ofproto *p = ofconn_get_ofproto(ofconn);
1561 struct ofp_packet_out *opo;
1562 struct ofpbuf payload, *buffer;
1563 union ofp_action *ofp_actions;
1564 struct ofpbuf request;
1566 size_t n_ofp_actions;
1570 COVERAGE_INC(ofproto_packet_out);
1572 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1577 /* Get ofp_packet_out. */
1578 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1579 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1582 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1583 &ofp_actions, &n_ofp_actions);
1589 if (opo->buffer_id != htonl(UINT32_MAX)) {
1590 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1592 if (error || !buffer) {
1601 /* Send out packet. */
1602 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1603 error = p->ofproto_class->packet_out(p, &payload, &flow,
1604 ofp_actions, n_ofp_actions);
1605 ofpbuf_delete(buffer);
1611 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1613 ovs_be32 old_config = port->opp.config;
1615 mask &= config ^ port->opp.config;
1616 if (mask & htonl(OFPPC_PORT_DOWN)) {
1617 if (config & htonl(OFPPC_PORT_DOWN)) {
1618 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1620 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1624 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1625 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1626 OFPPC_NO_PACKET_IN));
1627 if (port->opp.config != old_config) {
1628 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1633 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1635 struct ofproto *p = ofconn_get_ofproto(ofconn);
1636 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1637 struct ofport *port;
1640 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1645 port = ofproto_get_port(p, ntohs(opm->port_no));
1647 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1648 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1649 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1651 update_port_config(port, opm->config, opm->mask);
1652 if (opm->advertise) {
1653 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1660 handle_desc_stats_request(struct ofconn *ofconn,
1661 const struct ofp_stats_msg *request)
1663 struct ofproto *p = ofconn_get_ofproto(ofconn);
1664 struct ofp_desc_stats *ods;
1667 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1668 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1669 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1670 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1671 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1672 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1673 ofconn_send_reply(ofconn, msg);
1679 handle_table_stats_request(struct ofconn *ofconn,
1680 const struct ofp_stats_msg *request)
1682 struct ofproto *p = ofconn_get_ofproto(ofconn);
1683 struct ofp_table_stats *ots;
1687 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1689 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1690 for (i = 0; i < p->n_tables; i++) {
1691 ots[i].table_id = i;
1692 sprintf(ots[i].name, "table%zu", i);
1693 ots[i].wildcards = htonl(OFPFW_ALL);
1694 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1695 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1698 p->ofproto_class->get_tables(p, ots);
1700 ofconn_send_reply(ofconn, msg);
1705 append_port_stat(struct ofport *port, struct list *replies)
1707 struct netdev_stats stats;
1708 struct ofp_port_stats *ops;
1710 /* Intentionally ignore return value, since errors will set
1711 * 'stats' to all-1s, which is correct for OpenFlow, and
1712 * netdev_get_stats() will log errors. */
1713 netdev_get_stats(port->netdev, &stats);
1715 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1716 ops->port_no = port->opp.port_no;
1717 memset(ops->pad, 0, sizeof ops->pad);
1718 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1719 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1720 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1721 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1722 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1723 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1724 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1725 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1726 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1727 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1728 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1729 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1733 handle_port_stats_request(struct ofconn *ofconn,
1734 const struct ofp_port_stats_request *psr)
1736 struct ofproto *p = ofconn_get_ofproto(ofconn);
1737 struct ofport *port;
1738 struct list replies;
1740 ofputil_start_stats_reply(&psr->osm, &replies);
1741 if (psr->port_no != htons(OFPP_NONE)) {
1742 port = ofproto_get_port(p, ntohs(psr->port_no));
1744 append_port_stat(port, &replies);
1747 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1748 append_port_stat(port, &replies);
1752 ofconn_send_replies(ofconn, &replies);
1757 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1759 long long int msecs = time_msec() - start;
1760 *sec = msecs / 1000;
1761 *nsec = (msecs % 1000) * (1000 * 1000);
1764 static struct classifier *
1765 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1767 if (table_id == 0xff) {
1768 return &ofproto->tables[0];
1769 } else if (table_id < ofproto->n_tables) {
1770 return &ofproto->tables[table_id];
1772 /* It would probably be better to reply with an error but there doesn't
1773 * seem to be any appropriate value, so that might just be
1775 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1781 static struct classifier *
1782 next_matching_table(struct ofproto *ofproto,
1783 struct classifier *cls, uint8_t table_id)
1785 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1790 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1793 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1796 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1797 * only once, for that table.
1799 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1800 * and does not enter the loop at all.
1802 * All parameters are evaluated multiple times.
1804 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1805 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1807 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1809 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1810 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1811 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1814 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1815 * to 'out_port' are included.
1817 * Hidden rules are always omitted.
1819 * Returns 0 on success, otherwise an OpenFlow error code. */
1821 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1822 const struct cls_rule *match, uint16_t out_port,
1825 struct classifier *cls;
1828 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1829 struct cls_cursor cursor;
1832 cls_cursor_init(&cursor, cls, match);
1833 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1834 if (rule->pending) {
1835 return OFPROTO_POSTPONE;
1837 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1838 list_push_back(rules, &rule->ofproto_node);
1845 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1846 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1847 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1850 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1851 * to 'out_port' are included.
1853 * Hidden rules are always omitted.
1855 * Returns 0 on success, otherwise an OpenFlow error code. */
1857 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1858 const struct cls_rule *match, uint16_t out_port,
1861 struct classifier *cls;
1864 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1867 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1869 if (rule->pending) {
1870 return OFPROTO_POSTPONE;
1872 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1873 list_push_back(rules, &rule->ofproto_node);
1881 handle_flow_stats_request(struct ofconn *ofconn,
1882 const struct ofp_stats_msg *osm)
1884 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1885 struct flow_stats_request fsr;
1886 struct list replies;
1891 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1896 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1897 fsr.out_port, &rules);
1902 ofputil_start_stats_reply(osm, &replies);
1903 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1904 struct ofputil_flow_stats fs;
1907 fs.cookie = rule->flow_cookie;
1908 fs.table_id = rule->table_id;
1909 calc_flow_duration__(rule->created, &fs.duration_sec,
1911 fs.idle_timeout = rule->idle_timeout;
1912 fs.hard_timeout = rule->hard_timeout;
1913 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1915 fs.actions = rule->actions;
1916 fs.n_actions = rule->n_actions;
1917 ofputil_append_flow_stats_reply(&fs, &replies);
1919 ofconn_send_replies(ofconn, &replies);
1925 flow_stats_ds(struct rule *rule, struct ds *results)
1927 uint64_t packet_count, byte_count;
1929 rule->ofproto->ofproto_class->rule_get_stats(rule,
1930 &packet_count, &byte_count);
1932 if (rule->table_id != 0) {
1933 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1935 ds_put_format(results, "duration=%llds, ",
1936 (time_msec() - rule->created) / 1000);
1937 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1938 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1939 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1940 cls_rule_format(&rule->cr, results);
1941 ds_put_char(results, ',');
1942 if (rule->n_actions > 0) {
1943 ofp_print_actions(results, rule->actions, rule->n_actions);
1945 ds_put_cstr(results, "drop");
1947 ds_put_cstr(results, "\n");
1950 /* Adds a pretty-printed description of all flows to 'results', including
1951 * hidden flows (e.g., set up by in-band control). */
1953 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1955 struct classifier *cls;
1957 for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
1958 struct cls_cursor cursor;
1961 cls_cursor_init(&cursor, cls, NULL);
1962 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1963 flow_stats_ds(rule, results);
1968 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1969 * '*engine_type' and '*engine_id', respectively. */
1971 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1972 uint8_t *engine_type, uint8_t *engine_id)
1974 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1977 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1978 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1979 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
1981 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
1983 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1984 return (ofport && ofproto->ofproto_class->get_cfm_fault
1985 ? ofproto->ofproto_class->get_cfm_fault(ofport)
1990 handle_aggregate_stats_request(struct ofconn *ofconn,
1991 const struct ofp_stats_msg *osm)
1993 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1994 struct flow_stats_request request;
1995 struct ofputil_aggregate_stats stats;
1996 bool unknown_packets, unknown_bytes;
1997 struct ofpbuf *reply;
2002 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2007 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2008 request.out_port, &rules);
2013 memset(&stats, 0, sizeof stats);
2014 unknown_packets = unknown_bytes = false;
2015 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2016 uint64_t packet_count;
2017 uint64_t byte_count;
2019 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2022 if (packet_count == UINT64_MAX) {
2023 unknown_packets = true;
2025 stats.packet_count += packet_count;
2028 if (byte_count == UINT64_MAX) {
2029 unknown_bytes = true;
2031 stats.byte_count += byte_count;
2036 if (unknown_packets) {
2037 stats.packet_count = UINT64_MAX;
2039 if (unknown_bytes) {
2040 stats.byte_count = UINT64_MAX;
2043 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2044 ofconn_send_reply(ofconn, reply);
2049 struct queue_stats_cbdata {
2050 struct ofport *ofport;
2051 struct list replies;
2055 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2056 const struct netdev_queue_stats *stats)
2058 struct ofp_queue_stats *reply;
2060 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2061 reply->port_no = cbdata->ofport->opp.port_no;
2062 memset(reply->pad, 0, sizeof reply->pad);
2063 reply->queue_id = htonl(queue_id);
2064 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2065 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2066 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2070 handle_queue_stats_dump_cb(uint32_t queue_id,
2071 struct netdev_queue_stats *stats,
2074 struct queue_stats_cbdata *cbdata = cbdata_;
2076 put_queue_stats(cbdata, queue_id, stats);
2080 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2081 struct queue_stats_cbdata *cbdata)
2083 cbdata->ofport = port;
2084 if (queue_id == OFPQ_ALL) {
2085 netdev_dump_queue_stats(port->netdev,
2086 handle_queue_stats_dump_cb, cbdata);
2088 struct netdev_queue_stats stats;
2090 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2091 put_queue_stats(cbdata, queue_id, &stats);
2097 handle_queue_stats_request(struct ofconn *ofconn,
2098 const struct ofp_queue_stats_request *qsr)
2100 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2101 struct queue_stats_cbdata cbdata;
2102 struct ofport *port;
2103 unsigned int port_no;
2106 COVERAGE_INC(ofproto_queue_req);
2108 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2110 port_no = ntohs(qsr->port_no);
2111 queue_id = ntohl(qsr->queue_id);
2112 if (port_no == OFPP_ALL) {
2113 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2114 handle_queue_stats_for_port(port, queue_id, &cbdata);
2116 } else if (port_no < OFPP_MAX) {
2117 port = ofproto_get_port(ofproto, port_no);
2119 handle_queue_stats_for_port(port, queue_id, &cbdata);
2122 ofpbuf_list_delete(&cbdata.replies);
2123 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2125 ofconn_send_replies(ofconn, &cbdata.replies);
2131 is_flow_deletion_pending(const struct ofproto *ofproto,
2132 const struct cls_rule *cls_rule,
2135 if (!hmap_is_empty(&ofproto->deletions)) {
2136 struct ofoperation *op;
2138 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2139 cls_rule_hash(cls_rule, table_id),
2140 &ofproto->deletions) {
2141 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2150 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2151 * in which no matching flow already exists in the flow table.
2153 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2154 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2155 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2157 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2160 add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
2161 const struct ofp_header *request)
2163 struct classifier *table;
2164 struct ofopgroup *group;
2165 struct rule *victim;
2169 /* Check for overlap, if requested. */
2170 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2171 struct classifier *cls;
2173 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2174 if (classifier_rule_overlaps(cls, &fm->cr)) {
2175 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2181 if (fm->table_id == 0xff) {
2183 if (ofproto->n_tables > 1) {
2184 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2189 assert(table_id < ofproto->n_tables);
2190 table = &ofproto->tables[table_id];
2192 table = &ofproto->tables[0];
2194 } else if (fm->table_id < ofproto->n_tables) {
2195 table = &ofproto->tables[fm->table_id];
2197 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2200 /* Serialize against pending deletion. */
2201 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2202 return OFPROTO_POSTPONE;
2205 /* Allocate new rule. */
2206 rule = ofproto->ofproto_class->rule_alloc();
2208 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2209 ofproto->name, strerror(error));
2212 rule->ofproto = ofproto;
2214 rule->pending = NULL;
2215 rule->flow_cookie = fm->cookie;
2216 rule->created = time_msec();
2217 rule->idle_timeout = fm->idle_timeout;
2218 rule->hard_timeout = fm->hard_timeout;
2219 rule->table_id = table - ofproto->tables;
2220 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2221 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2222 rule->n_actions = fm->n_actions;
2224 /* Insert new rule. */
2225 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2226 if (victim && victim->pending) {
2227 error = OFPROTO_POSTPONE;
2230 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2231 : ofopgroup_create(ofproto));
2232 ofoperation_create(group, rule, OFOPERATION_ADD);
2233 rule->pending->victim = victim;
2235 error = ofproto->ofproto_class->rule_construct(rule);
2237 ofoperation_destroy(rule->pending);
2239 ofopgroup_submit(group);
2242 /* Back out if an error occurred. */
2245 classifier_replace(table, &victim->cr);
2247 classifier_remove(table, &rule->cr);
2249 ofproto_rule_destroy__(rule);
2254 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2256 /* Modifies the rules listed in 'rules', changing their actions to match those
2259 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2262 * Returns 0 on success, otherwise an OpenFlow error code. */
2264 modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
2265 const struct ofp_header *request, struct list *rules)
2267 struct ofopgroup *group;
2270 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2271 LIST_FOR_EACH (rule, ofproto_node, rules) {
2272 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2273 rule->actions, rule->n_actions)) {
2274 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2275 rule->pending->actions = rule->actions;
2276 rule->pending->n_actions = rule->n_actions;
2277 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2278 rule->n_actions = fm->n_actions;
2279 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2281 rule->flow_cookie = fm->cookie;
2283 ofopgroup_submit(group);
2288 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2289 * encoded by ofp_mkerr() on failure.
2291 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2294 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
2295 const struct ofp_header *request)
2297 struct ofproto *p = ofconn_get_ofproto(ofconn);
2301 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2302 return (error ? error
2303 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2304 : modify_flows__(ofconn, fm, request, &rules));
2307 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2308 * code as encoded by ofp_mkerr() on failure.
2310 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2313 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2314 const struct ofp_header *request)
2316 struct ofproto *p = ofconn_get_ofproto(ofconn);
2320 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2321 return (error ? error
2322 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2323 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2328 /* OFPFC_DELETE implementation. */
2330 /* Deletes the rules listed in 'rules'.
2332 * Returns 0 on success, otherwise an OpenFlow error code. */
2334 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2337 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2338 struct rule *rule, *next;
2339 struct ofopgroup *group;
2341 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2342 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2343 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2345 ofoperation_create(group, rule, OFOPERATION_DELETE);
2346 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2347 rule->ofproto->ofproto_class->rule_destruct(rule);
2349 ofopgroup_submit(group);
2354 /* Implements OFPFC_DELETE. */
2356 delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
2357 const struct ofp_header *request)
2359 struct ofproto *p = ofconn_get_ofproto(ofconn);
2363 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2365 return (error ? error
2366 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2370 /* Implements OFPFC_DELETE_STRICT. */
2372 delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2373 const struct ofp_header *request)
2375 struct ofproto *p = ofconn_get_ofproto(ofconn);
2379 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2381 return (error ? error
2382 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2388 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2390 struct ofputil_flow_removed fr;
2392 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2397 fr.cookie = rule->flow_cookie;
2399 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2400 fr.idle_timeout = rule->idle_timeout;
2401 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2404 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2407 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2408 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2411 * ofproto implementation ->run() functions should use this function to expire
2412 * OpenFlow flows. */
2414 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2416 struct ofproto *ofproto = rule->ofproto;
2417 struct ofopgroup *group;
2419 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2421 ofproto_rule_send_removed(rule, reason);
2423 group = ofopgroup_create(ofproto);
2424 ofoperation_create(group, rule, OFOPERATION_DELETE);
2425 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2426 rule->ofproto->ofproto_class->rule_destruct(rule);
2427 ofopgroup_submit(group);
2431 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2433 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2437 error = reject_slave_controller(ofconn, "flow_mod");
2442 if (list_size(&ofproto->pending) >= 50) {
2443 return OFPROTO_POSTPONE;
2446 error = ofputil_decode_flow_mod(&fm, oh,
2447 ofconn_get_flow_mod_table_id(ofconn));
2452 /* We do not support the emergency flow cache. It will hopefully get
2453 * dropped from OpenFlow in the near future. */
2454 if (fm.flags & OFPFF_EMERG) {
2455 /* There isn't a good fit for an error code, so just state that the
2456 * flow table is full. */
2457 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2460 switch (fm.command) {
2462 return add_flow(ofproto, ofconn, &fm, oh);
2465 return modify_flows_loose(ofconn, &fm, oh);
2467 case OFPFC_MODIFY_STRICT:
2468 return modify_flow_strict(ofconn, &fm, oh);
2471 return delete_flows_loose(ofconn, &fm, oh);
2473 case OFPFC_DELETE_STRICT:
2474 return delete_flow_strict(ofconn, &fm, oh);
2477 if (fm.command > 0xff) {
2478 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2479 "flow_mod_table_id extension is not enabled");
2481 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2486 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2488 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2489 struct nx_role_request *reply;
2493 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2494 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2495 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2498 role = ntohl(nrr->role);
2499 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2500 && role != NX_ROLE_SLAVE) {
2501 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2503 /* There's no good error code for this. */
2504 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2507 if (ofconn_get_role(ofconn) != role
2508 && ofconn_has_pending_opgroups(ofconn)) {
2509 return OFPROTO_POSTPONE;
2512 ofconn_set_role(ofconn, role);
2514 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2515 reply->role = htonl(role);
2516 ofconn_send_reply(ofconn, buf);
2522 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2523 const struct ofp_header *oh)
2525 const struct nxt_flow_mod_table_id *msg
2526 = (const struct nxt_flow_mod_table_id *) oh;
2528 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2533 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2535 const struct nxt_set_flow_format *msg
2536 = (const struct nxt_set_flow_format *) oh;
2539 format = ntohl(msg->format);
2540 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2541 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2544 if (format != ofconn_get_flow_format(ofconn)
2545 && ofconn_has_pending_opgroups(ofconn)) {
2546 /* Avoid sending async messages in surprising flow format. */
2547 return OFPROTO_POSTPONE;
2550 ofconn_set_flow_format(ofconn, format);
2555 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2557 struct ofp_header *ob;
2560 if (ofconn_has_pending_opgroups(ofconn)) {
2561 return OFPROTO_POSTPONE;
2564 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2565 ofconn_send_reply(ofconn, buf);
2570 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2572 const struct ofp_header *oh = msg->data;
2573 const struct ofputil_msg_type *type;
2576 error = ofputil_decode_msg_type(oh, &type);
2581 switch (ofputil_msg_type_code(type)) {
2582 /* OpenFlow requests. */
2583 case OFPUTIL_OFPT_ECHO_REQUEST:
2584 return handle_echo_request(ofconn, oh);
2586 case OFPUTIL_OFPT_FEATURES_REQUEST:
2587 return handle_features_request(ofconn, oh);
2589 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2590 return handle_get_config_request(ofconn, oh);
2592 case OFPUTIL_OFPT_SET_CONFIG:
2593 return handle_set_config(ofconn, msg->data);
2595 case OFPUTIL_OFPT_PACKET_OUT:
2596 return handle_packet_out(ofconn, oh);
2598 case OFPUTIL_OFPT_PORT_MOD:
2599 return handle_port_mod(ofconn, oh);
2601 case OFPUTIL_OFPT_FLOW_MOD:
2602 return handle_flow_mod(ofconn, oh);
2604 case OFPUTIL_OFPT_BARRIER_REQUEST:
2605 return handle_barrier_request(ofconn, oh);
2607 /* OpenFlow replies. */
2608 case OFPUTIL_OFPT_ECHO_REPLY:
2611 /* Nicira extension requests. */
2612 case OFPUTIL_NXT_ROLE_REQUEST:
2613 return handle_role_request(ofconn, oh);
2615 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2616 return handle_nxt_flow_mod_table_id(ofconn, oh);
2618 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2619 return handle_nxt_set_flow_format(ofconn, oh);
2621 case OFPUTIL_NXT_FLOW_MOD:
2622 return handle_flow_mod(ofconn, oh);
2624 /* Statistics requests. */
2625 case OFPUTIL_OFPST_DESC_REQUEST:
2626 return handle_desc_stats_request(ofconn, msg->data);
2628 case OFPUTIL_OFPST_FLOW_REQUEST:
2629 case OFPUTIL_NXST_FLOW_REQUEST:
2630 return handle_flow_stats_request(ofconn, msg->data);
2632 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2633 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2634 return handle_aggregate_stats_request(ofconn, msg->data);
2636 case OFPUTIL_OFPST_TABLE_REQUEST:
2637 return handle_table_stats_request(ofconn, msg->data);
2639 case OFPUTIL_OFPST_PORT_REQUEST:
2640 return handle_port_stats_request(ofconn, msg->data);
2642 case OFPUTIL_OFPST_QUEUE_REQUEST:
2643 return handle_queue_stats_request(ofconn, msg->data);
2645 case OFPUTIL_MSG_INVALID:
2646 case OFPUTIL_OFPT_HELLO:
2647 case OFPUTIL_OFPT_ERROR:
2648 case OFPUTIL_OFPT_FEATURES_REPLY:
2649 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2650 case OFPUTIL_OFPT_PACKET_IN:
2651 case OFPUTIL_OFPT_FLOW_REMOVED:
2652 case OFPUTIL_OFPT_PORT_STATUS:
2653 case OFPUTIL_OFPT_BARRIER_REPLY:
2654 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2655 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2656 case OFPUTIL_OFPST_DESC_REPLY:
2657 case OFPUTIL_OFPST_FLOW_REPLY:
2658 case OFPUTIL_OFPST_QUEUE_REPLY:
2659 case OFPUTIL_OFPST_PORT_REPLY:
2660 case OFPUTIL_OFPST_TABLE_REPLY:
2661 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2662 case OFPUTIL_NXT_ROLE_REPLY:
2663 case OFPUTIL_NXT_FLOW_REMOVED:
2664 case OFPUTIL_NXST_FLOW_REPLY:
2665 case OFPUTIL_NXST_AGGREGATE_REPLY:
2667 if (VLOG_IS_WARN_ENABLED()) {
2668 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2669 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2672 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2673 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2675 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2681 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2683 int error = handle_openflow__(ofconn, ofp_msg);
2684 if (error && error != OFPROTO_POSTPONE) {
2685 ofconn_send_error(ofconn, ofp_msg->data, error);
2687 COVERAGE_INC(ofproto_recv_openflow);
2688 return error != OFPROTO_POSTPONE;
2691 /* Asynchronous operations. */
2693 /* Creates and returns a new ofopgroup that is not associated with any
2694 * OpenFlow connection.
2696 * The caller should add operations to the returned group with
2697 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2698 static struct ofopgroup *
2699 ofopgroup_create(struct ofproto *ofproto)
2701 struct ofopgroup *group = xzalloc(sizeof *group);
2702 group->ofproto = ofproto;
2703 list_init(&group->ofproto_node);
2704 list_init(&group->ops);
2705 list_init(&group->ofconn_node);
2709 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2710 * the ofopgroup eventually fails, then the error reply will include 'request'.
2711 * If the ofopgroup eventually succeeds, then the packet with buffer id
2712 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2714 * The caller should add operations to the returned group with
2715 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2716 static struct ofopgroup *
2717 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2718 const struct ofp_header *request,
2721 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2722 size_t request_len = ntohs(request->length);
2724 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2725 group->ofconn = ofconn;
2726 group->request = xmemdup(request, MIN(request_len, 64));
2727 group->buffer_id = buffer_id;
2732 /* Submits 'group' for processing.
2734 * If 'group' contains no operations (e.g. none were ever added, or all of the
2735 * ones that were added completed synchronously), then it is destroyed
2736 * immediately. Otherwise it is added to the ofproto's list of pending
2739 ofopgroup_submit(struct ofopgroup *group)
2741 if (list_is_empty(&group->ops)) {
2742 ofopgroup_destroy(group);
2744 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2749 ofopgroup_destroy(struct ofopgroup *group)
2751 assert(list_is_empty(&group->ops));
2752 if (!list_is_empty(&group->ofproto_node)) {
2753 list_remove(&group->ofproto_node);
2755 if (!list_is_empty(&group->ofconn_node)) {
2756 list_remove(&group->ofconn_node);
2758 ofconn_send_error(group->ofconn, group->request, group->error);
2760 connmgr_retry(group->ofproto->connmgr);
2762 free(group->request);
2766 /* Initiates a new operation on 'rule', of the specified 'type', within
2767 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2769 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2770 enum ofoperation_type type)
2772 struct ofoperation *op;
2774 assert(!rule->pending);
2776 op = rule->pending = xzalloc(sizeof *op);
2778 list_push_back(&group->ops, &op->group_node);
2782 op->flow_cookie = rule->flow_cookie;
2784 if (type == OFOPERATION_DELETE) {
2785 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2786 cls_rule_hash(&rule->cr, rule->table_id));
2791 ofoperation_destroy(struct ofoperation *op)
2793 struct ofopgroup *group = op->group;
2796 op->rule->pending = NULL;
2798 if (op->type == OFOPERATION_DELETE) {
2799 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2801 list_remove(&op->group_node);
2805 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2806 ofopgroup_destroy(group);
2810 /* Indicates that 'op' completed with status 'error', which is either 0 to
2811 * indicate success or an OpenFlow error code (constructed with
2812 * e.g. ofp_mkerr()).
2814 * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
2815 * deletions are not allowed to fail.
2817 * Please see the large comment in ofproto/ofproto-provider.h titled
2818 * "Asynchronous Operation Support" for more information. */
2820 ofoperation_complete(struct ofoperation *op, int error)
2822 struct ofopgroup *group = op->group;
2823 struct rule *rule = op->rule;
2824 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2826 assert(rule->pending == op);
2827 assert(op->status < 0);
2832 && op->type != OFOPERATION_DELETE
2834 && group->buffer_id != UINT32_MAX
2835 && list_is_singleton(&op->group_node)) {
2836 struct ofpbuf *packet;
2839 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2843 error = rule_execute(rule, in_port, packet);
2846 if (!group->error) {
2847 group->error = error;
2851 case OFOPERATION_ADD:
2854 ofproto_rule_destroy__(op->victim);
2858 classifier_replace(table, &op->victim->cr);
2861 classifier_remove(table, &rule->cr);
2863 ofproto_rule_destroy__(rule);
2868 case OFOPERATION_DELETE:
2870 ofproto_rule_destroy__(rule);
2874 case OFOPERATION_MODIFY:
2876 free(rule->actions);
2877 rule->actions = op->actions;
2878 rule->n_actions = op->n_actions;
2886 ofoperation_destroy(op);
2890 ofoperation_get_victim(struct ofoperation *op)
2892 assert(op->type == OFOPERATION_ADD);
2897 pick_datapath_id(const struct ofproto *ofproto)
2899 const struct ofport *port;
2901 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2903 uint8_t ea[ETH_ADDR_LEN];
2906 error = netdev_get_etheraddr(port->netdev, ea);
2908 return eth_addr_to_uint64(ea);
2910 VLOG_WARN("could not get MAC address for %s (%s)",
2911 netdev_get_name(port->netdev), strerror(error));
2913 return ofproto->fallback_dpid;
2917 pick_fallback_dpid(void)
2919 uint8_t ea[ETH_ADDR_LEN];
2920 eth_addr_nicira_random(ea);
2921 return eth_addr_to_uint64(ea);
2924 /* unixctl commands. */
2927 ofproto_lookup(const char *name)
2929 struct ofproto *ofproto;
2931 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2933 if (!strcmp(ofproto->name, name)) {
2941 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2942 void *aux OVS_UNUSED)
2944 struct ofproto *ofproto;
2948 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2949 ds_put_format(&results, "%s\n", ofproto->name);
2951 unixctl_command_reply(conn, 200, ds_cstr(&results));
2952 ds_destroy(&results);
2956 ofproto_unixctl_init(void)
2958 static bool registered;
2964 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);