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 *,
139 struct ofputil_flow_mod *, 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 classifier *table;
293 struct ofproto *ofproto;
299 ofproto_initialize();
300 ofproto_unixctl_init();
302 datapath_type = ofproto_normalize_type(datapath_type);
303 class = ofproto_class_find__(datapath_type);
305 VLOG_WARN("could not create datapath %s of unknown type %s",
306 datapath_name, datapath_type);
310 ofproto = class->alloc();
312 VLOG_ERR("failed to allocate datapath %s of type %s",
313 datapath_name, datapath_type);
318 memset(ofproto, 0, sizeof *ofproto);
319 ofproto->ofproto_class = class;
320 ofproto->name = xstrdup(datapath_name);
321 ofproto->type = xstrdup(datapath_type);
322 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
323 hash_string(ofproto->name, 0));
324 ofproto->datapath_id = 0;
325 ofproto_set_flow_eviction_threshold(ofproto,
326 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
327 ofproto->forward_bpdu = false;
328 ofproto->fallback_dpid = pick_fallback_dpid();
329 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
330 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
331 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
332 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
333 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
334 hmap_init(&ofproto->ports);
335 shash_init(&ofproto->port_by_name);
336 ofproto->tables = NULL;
337 ofproto->n_tables = 0;
338 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
339 ofproto->state = S_OPENFLOW;
340 list_init(&ofproto->pending);
341 ofproto->n_pending = 0;
342 hmap_init(&ofproto->deletions);
344 error = ofproto->ofproto_class->construct(ofproto, &n_tables);
346 VLOG_ERR("failed to open datapath %s: %s",
347 datapath_name, strerror(error));
348 ofproto_destroy__(ofproto);
352 assert(n_tables >= 1 && n_tables <= 255);
353 ofproto->n_tables = n_tables;
354 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
355 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
356 classifier_init(table);
359 ofproto->datapath_id = pick_datapath_id(ofproto);
360 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
368 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
370 uint64_t old_dpid = p->datapath_id;
371 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
372 if (p->datapath_id != old_dpid) {
373 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
375 /* Force all active connections to reconnect, since there is no way to
376 * notify a controller that the datapath ID has changed. */
377 ofproto_reconnect_controllers(p);
382 ofproto_set_controllers(struct ofproto *p,
383 const struct ofproto_controller *controllers,
384 size_t n_controllers)
386 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
390 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
392 connmgr_set_fail_mode(p->connmgr, fail_mode);
395 /* Drops the connections between 'ofproto' and all of its controllers, forcing
396 * them to reconnect. */
398 ofproto_reconnect_controllers(struct ofproto *ofproto)
400 connmgr_reconnect(ofproto->connmgr);
403 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
404 * in-band control should guarantee access, in the same way that in-band
405 * control guarantees access to OpenFlow controllers. */
407 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
408 const struct sockaddr_in *extras, size_t n)
410 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
413 /* Sets the OpenFlow queue used by flows set up by in-band control on
414 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
415 * flows will use the default queue. */
417 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
419 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
422 /* Sets the number of flows at which eviction from the kernel flow table
425 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
427 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
428 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
430 ofproto->flow_eviction_threshold = threshold;
434 /* If forward_bpdu is true, the NORMAL action will forward frames with
435 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
436 * the NORMAL action will drop these frames. */
438 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
440 bool old_val = ofproto->forward_bpdu;
441 ofproto->forward_bpdu = forward_bpdu;
442 if (old_val != ofproto->forward_bpdu) {
443 if (ofproto->ofproto_class->forward_bpdu_changed) {
444 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
450 ofproto_set_desc(struct ofproto *p,
451 const char *mfr_desc, const char *hw_desc,
452 const char *sw_desc, const char *serial_desc,
455 struct ofp_desc_stats *ods;
458 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
459 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
460 sizeof ods->mfr_desc);
463 p->mfr_desc = xstrdup(mfr_desc);
466 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
467 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
468 sizeof ods->hw_desc);
471 p->hw_desc = xstrdup(hw_desc);
474 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
475 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
476 sizeof ods->sw_desc);
479 p->sw_desc = xstrdup(sw_desc);
482 if (strlen(serial_desc) >= sizeof ods->serial_num) {
483 VLOG_WARN("truncating serial_desc, must be less than %zu "
485 sizeof ods->serial_num);
487 free(p->serial_desc);
488 p->serial_desc = xstrdup(serial_desc);
491 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
492 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
493 sizeof ods->dp_desc);
496 p->dp_desc = xstrdup(dp_desc);
501 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
503 return connmgr_set_snoops(ofproto->connmgr, snoops);
507 ofproto_set_netflow(struct ofproto *ofproto,
508 const struct netflow_options *nf_options)
510 if (nf_options && sset_is_empty(&nf_options->collectors)) {
514 if (ofproto->ofproto_class->set_netflow) {
515 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
517 return nf_options ? EOPNOTSUPP : 0;
522 ofproto_set_sflow(struct ofproto *ofproto,
523 const struct ofproto_sflow_options *oso)
525 if (oso && sset_is_empty(&oso->targets)) {
529 if (ofproto->ofproto_class->set_sflow) {
530 return ofproto->ofproto_class->set_sflow(ofproto, oso);
532 return oso ? EOPNOTSUPP : 0;
536 /* Connectivity Fault Management configuration. */
538 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
540 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
542 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
543 if (ofport && ofproto->ofproto_class->set_cfm) {
544 ofproto->ofproto_class->set_cfm(ofport, NULL);
548 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
549 * basic configuration from the configuration members in 'cfm', and the remote
550 * maintenance point ID from remote_mpid. Ignores the statistics members of
553 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
555 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
556 const struct cfm_settings *s)
558 struct ofport *ofport;
561 ofport = ofproto_get_port(ofproto, ofp_port);
563 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
564 ofproto->name, ofp_port);
568 /* XXX: For configuration simplicity, we only support one remote_mpid
569 * outside of the CFM module. It's not clear if this is the correct long
570 * term solution or not. */
571 error = (ofproto->ofproto_class->set_cfm
572 ? ofproto->ofproto_class->set_cfm(ofport, s)
575 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
576 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
581 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
582 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
583 * 0 if LACP partner information is not current (generally indicating a
584 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
586 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
588 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
589 return (ofport && ofproto->ofproto_class->port_is_lacp_current
590 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
596 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
597 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
598 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
599 * configuration plus, if there is more than one slave, a bonding
602 * If 'aux' is already registered then this function updates its configuration
603 * to 's'. Otherwise, this function registers a new bundle.
605 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
608 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
609 const struct ofproto_bundle_settings *s)
611 return (ofproto->ofproto_class->bundle_set
612 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
616 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
617 * If no such bundle has been registered, this has no effect. */
619 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
621 return ofproto_bundle_register(ofproto, aux, NULL);
625 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
626 * If 'aux' is already registered then this function updates its configuration
627 * to 's'. Otherwise, this function registers a new mirror.
629 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
632 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
633 const struct ofproto_mirror_settings *s)
635 return (ofproto->ofproto_class->mirror_set
636 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
640 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
641 * If no mirror has been registered, this has no effect. */
643 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
645 return ofproto_mirror_register(ofproto, aux, NULL);
648 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
649 * which all packets are flooded, instead of using MAC learning. If
650 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
652 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
655 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
657 return (ofproto->ofproto_class->set_flood_vlans
658 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
662 /* Returns true if 'aux' is a registered bundle that is currently in use as the
663 * output for a mirror. */
665 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
667 return (ofproto->ofproto_class->is_mirror_output_bundle
668 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
673 ofproto_has_snoops(const struct ofproto *ofproto)
675 return connmgr_has_snoops(ofproto->connmgr);
679 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
681 connmgr_get_snoops(ofproto->connmgr, snoops);
685 ofproto_flush__(struct ofproto *ofproto)
687 struct classifier *table;
688 struct ofopgroup *group;
690 if (ofproto->ofproto_class->flush) {
691 ofproto->ofproto_class->flush(ofproto);
694 group = ofopgroup_create(ofproto);
695 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
696 struct rule *rule, *next_rule;
697 struct cls_cursor cursor;
699 cls_cursor_init(&cursor, table, NULL);
700 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
701 if (!rule->pending) {
702 ofoperation_create(group, rule, OFOPERATION_DELETE);
703 classifier_remove(table, &rule->cr);
704 ofproto->ofproto_class->rule_destruct(rule);
708 ofopgroup_submit(group);
712 ofproto_destroy__(struct ofproto *ofproto)
714 struct classifier *table;
716 assert(list_is_empty(&ofproto->pending));
717 assert(!ofproto->n_pending);
719 connmgr_destroy(ofproto->connmgr);
721 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
724 free(ofproto->mfr_desc);
725 free(ofproto->hw_desc);
726 free(ofproto->sw_desc);
727 free(ofproto->serial_desc);
728 free(ofproto->dp_desc);
729 hmap_destroy(&ofproto->ports);
730 shash_destroy(&ofproto->port_by_name);
732 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
733 assert(classifier_is_empty(table));
734 classifier_destroy(table);
736 free(ofproto->tables);
738 hmap_destroy(&ofproto->deletions);
740 ofproto->ofproto_class->dealloc(ofproto);
744 ofproto_destroy(struct ofproto *p)
746 struct ofport *ofport, *next_ofport;
753 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
754 ofport_destroy(ofport);
757 p->ofproto_class->destruct(p);
758 ofproto_destroy__(p);
761 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
762 * kernel datapath, for example, this destroys the datapath in the kernel, and
763 * with the netdev-based datapath, it tears down the data structures that
764 * represent the datapath.
766 * The datapath should not be currently open as an ofproto. */
768 ofproto_delete(const char *name, const char *type)
770 const struct ofproto_class *class = ofproto_class_find__(type);
771 return (!class ? EAFNOSUPPORT
772 : !class->del ? EACCES
773 : class->del(type, name));
777 process_port_change(struct ofproto *ofproto, int error, char *devname)
779 if (error == ENOBUFS) {
780 reinit_ports(ofproto);
782 update_port(ofproto, devname);
788 ofproto_run(struct ofproto *p)
790 struct ofport *ofport;
794 error = p->ofproto_class->run(p);
795 if (error == ENODEV) {
796 /* Someone destroyed the datapath behind our back. The caller
797 * better destroy us and give up, because we're just going to
798 * spin from here on out. */
799 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
800 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
805 if (p->ofproto_class->port_poll) {
806 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
807 process_port_change(p, error, devname);
811 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
812 unsigned int change_seq = netdev_change_seq(ofport->netdev);
813 if (ofport->change_seq != change_seq) {
814 ofport->change_seq = change_seq;
815 update_port(p, netdev_get_name(ofport->netdev));
822 connmgr_run(p->connmgr, handle_openflow);
826 connmgr_run(p->connmgr, NULL);
828 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
829 connmgr_flushed(p->connmgr);
830 p->state = S_OPENFLOW;
842 ofproto_wait(struct ofproto *p)
844 struct ofport *ofport;
846 p->ofproto_class->wait(p);
847 if (p->ofproto_class->port_poll_wait) {
848 p->ofproto_class->port_poll_wait(p);
851 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
852 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
853 poll_immediate_wake();
859 connmgr_wait(p->connmgr, true);
863 connmgr_wait(p->connmgr, false);
864 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
865 poll_immediate_wake();
872 ofproto_is_alive(const struct ofproto *p)
874 return connmgr_has_controllers(p->connmgr);
878 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
881 connmgr_get_controller_info(ofproto->connmgr, info);
885 ofproto_free_ofproto_controller_info(struct shash *info)
887 connmgr_free_controller_info(info);
890 /* Makes a deep copy of 'old' into 'port'. */
892 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
894 port->name = xstrdup(old->name);
895 port->type = xstrdup(old->type);
896 port->ofp_port = old->ofp_port;
899 /* Frees memory allocated to members of 'ofproto_port'.
901 * Do not call this function on an ofproto_port obtained from
902 * ofproto_port_dump_next(): that function retains ownership of the data in the
905 ofproto_port_destroy(struct ofproto_port *ofproto_port)
907 free(ofproto_port->name);
908 free(ofproto_port->type);
911 /* Initializes 'dump' to begin dumping the ports in an ofproto.
913 * This function provides no status indication. An error status for the entire
914 * dump operation is provided when it is completed by calling
915 * ofproto_port_dump_done().
918 ofproto_port_dump_start(struct ofproto_port_dump *dump,
919 const struct ofproto *ofproto)
921 dump->ofproto = ofproto;
922 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
926 /* Attempts to retrieve another port from 'dump', which must have been created
927 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
928 * 'port' and returns true. On failure, returns false.
930 * Failure might indicate an actual error or merely that the last port has been
931 * dumped. An error status for the entire dump operation is provided when it
932 * is completed by calling ofproto_port_dump_done().
934 * The ofproto owns the data stored in 'port'. It will remain valid until at
935 * least the next time 'dump' is passed to ofproto_port_dump_next() or
936 * ofproto_port_dump_done(). */
938 ofproto_port_dump_next(struct ofproto_port_dump *dump,
939 struct ofproto_port *port)
941 const struct ofproto *ofproto = dump->ofproto;
947 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
950 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
956 /* Completes port table dump operation 'dump', which must have been created
957 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
958 * error-free, otherwise a positive errno value describing the problem. */
960 ofproto_port_dump_done(struct ofproto_port_dump *dump)
962 const struct ofproto *ofproto = dump->ofproto;
964 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
967 return dump->error == EOF ? 0 : dump->error;
970 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
971 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
972 * is non-null). On failure, returns a positive errno value and sets
973 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
975 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
981 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
983 update_port(ofproto, netdev_get_name(netdev));
986 *ofp_portp = error ? OFPP_NONE : ofp_port;
991 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
992 * initializes '*port' appropriately; on failure, returns a positive errno
995 * The caller owns the data in 'ofproto_port' and must free it with
996 * ofproto_port_destroy() when it is no longer needed. */
998 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
999 struct ofproto_port *port)
1003 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1005 memset(port, 0, sizeof *port);
1010 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1011 * Returns 0 if successful, otherwise a positive errno. */
1013 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1015 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1016 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1019 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1020 if (!error && ofport) {
1021 /* 'name' is the netdev's name and update_port() is going to close the
1022 * netdev. Just in case update_port() refers to 'name' after it
1023 * destroys 'ofport', make a copy of it around the update_port()
1025 char *devname = xstrdup(name);
1026 update_port(ofproto, devname);
1032 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1033 * performs the 'n_actions' actions in 'actions'. The new flow will not
1036 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1037 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1038 * controllers; otherwise, it will be hidden.
1040 * The caller retains ownership of 'cls_rule' and 'actions'.
1042 * This is a helper function for in-band control and fail-open. */
1044 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1045 const union ofp_action *actions, size_t n_actions)
1047 const struct rule *rule;
1049 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1050 &ofproto->tables[0], cls_rule));
1051 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1052 actions, n_actions)) {
1053 struct ofputil_flow_mod fm;
1055 memset(&fm, 0, sizeof fm);
1057 fm.buffer_id = UINT32_MAX;
1058 fm.actions = (union ofp_action *) actions;
1059 fm.n_actions = n_actions;
1060 add_flow(ofproto, NULL, &fm, NULL);
1064 /* Searches for a rule with matching criteria exactly equal to 'target' in
1065 * ofproto's table 0 and, if it finds one, deletes it.
1067 * This is a helper function for in-band control and fail-open. */
1069 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1073 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1074 &ofproto->tables[0], target));
1076 /* No such rule -> success. */
1078 } else if (rule->pending) {
1079 /* An operation on the rule is already pending -> failure.
1080 * Caller must retry later if it's important. */
1083 /* Initiate deletion -> success. */
1084 struct ofopgroup *group = ofopgroup_create(ofproto);
1085 ofoperation_create(group, rule, OFOPERATION_DELETE);
1086 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1087 rule->ofproto->ofproto_class->rule_destruct(rule);
1088 ofopgroup_submit(group);
1094 /* Starts the process of deleting all of the flows from all of ofproto's flow
1095 * tables and then reintroducing the flows required by in-band control and
1096 * fail-open. The process will complete in a later call to ofproto_run(). */
1098 ofproto_flush_flows(struct ofproto *ofproto)
1100 COVERAGE_INC(ofproto_flush);
1101 ofproto->state = S_FLUSH;
1105 reinit_ports(struct ofproto *p)
1107 struct ofproto_port_dump dump;
1108 struct sset devnames;
1109 struct ofport *ofport;
1110 struct ofproto_port ofproto_port;
1111 const char *devname;
1113 COVERAGE_INC(ofproto_reinit_ports);
1115 sset_init(&devnames);
1116 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1117 sset_add(&devnames, netdev_get_name(ofport->netdev));
1119 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1120 sset_add(&devnames, ofproto_port.name);
1123 SSET_FOR_EACH (devname, &devnames) {
1124 update_port(p, devname);
1126 sset_destroy(&devnames);
1129 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1130 * netdev cannot be opened. On success, also fills in 'opp'. */
1131 static struct netdev *
1132 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1134 uint32_t curr, advertised, supported, peer;
1135 enum netdev_flags flags;
1136 struct netdev *netdev;
1139 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1141 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1142 "cannot be opened (%s)",
1143 ofproto_port->name, ofproto_port->ofp_port,
1144 ofproto_port->name, strerror(error));
1148 netdev_get_flags(netdev, &flags);
1149 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1151 opp->port_no = htons(ofproto_port->ofp_port);
1152 netdev_get_etheraddr(netdev, opp->hw_addr);
1153 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1154 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1155 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1156 opp->curr = htonl(curr);
1157 opp->advertised = htonl(advertised);
1158 opp->supported = htonl(supported);
1159 opp->peer = htonl(peer);
1164 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1165 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1168 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1170 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1171 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1172 && a->state == b->state
1173 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1174 && a->curr == b->curr
1175 && a->advertised == b->advertised
1176 && a->supported == b->supported
1177 && a->peer == b->peer);
1180 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1181 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1182 * one with the same name or port number). */
1184 ofport_install(struct ofproto *p,
1185 struct netdev *netdev, const struct ofp_phy_port *opp)
1187 const char *netdev_name = netdev_get_name(netdev);
1188 struct ofport *ofport;
1191 /* Create ofport. */
1192 ofport = p->ofproto_class->port_alloc();
1197 ofport->ofproto = p;
1198 ofport->netdev = netdev;
1199 ofport->change_seq = netdev_change_seq(netdev);
1201 ofport->ofp_port = ntohs(opp->port_no);
1203 /* Add port to 'p'. */
1204 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1205 shash_add(&p->port_by_name, netdev_name, ofport);
1207 /* Let the ofproto_class initialize its private data. */
1208 error = p->ofproto_class->port_construct(ofport);
1212 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1216 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1217 p->name, netdev_name, strerror(error));
1219 ofport_destroy__(ofport);
1221 netdev_close(netdev);
1225 /* Removes 'ofport' from 'p' and destroys it. */
1227 ofport_remove(struct ofport *ofport)
1229 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1231 ofport_destroy(ofport);
1234 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1237 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1239 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1241 ofport_remove(port);
1245 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1247 * Does not handle a name or port number change. The caller must implement
1248 * such a change as a delete followed by an add. */
1250 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1252 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1253 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1254 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1255 port->opp.state = opp->state;
1256 port->opp.curr = opp->curr;
1257 port->opp.advertised = opp->advertised;
1258 port->opp.supported = opp->supported;
1259 port->opp.peer = opp->peer;
1261 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1265 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1267 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1269 if (port->ofproto->ofproto_class->set_cfm) {
1270 port->ofproto->ofproto_class->set_cfm(port, NULL);
1272 if (port->ofproto->ofproto_class->bundle_remove) {
1273 port->ofproto->ofproto_class->bundle_remove(port);
1279 ofport_destroy__(struct ofport *port)
1281 struct ofproto *ofproto = port->ofproto;
1282 const char *name = netdev_get_name(port->netdev);
1284 hmap_remove(&ofproto->ports, &port->hmap_node);
1285 shash_delete(&ofproto->port_by_name,
1286 shash_find(&ofproto->port_by_name, name));
1288 netdev_close(port->netdev);
1289 ofproto->ofproto_class->port_dealloc(port);
1293 ofport_destroy(struct ofport *port)
1296 port->ofproto->ofproto_class->port_destruct(port);
1297 ofport_destroy__(port);
1302 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1304 struct ofport *port;
1306 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1307 hash_int(ofp_port, 0), &ofproto->ports) {
1308 if (port->ofp_port == ofp_port) {
1316 update_port(struct ofproto *ofproto, const char *name)
1318 struct ofproto_port ofproto_port;
1319 struct ofp_phy_port opp;
1320 struct netdev *netdev;
1321 struct ofport *port;
1323 COVERAGE_INC(ofproto_update_port);
1325 /* Fetch 'name''s location and properties from the datapath. */
1326 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1327 ? ofport_open(&ofproto_port, &opp)
1330 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1331 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1332 struct netdev *old_netdev = port->netdev;
1334 /* 'name' hasn't changed location. Any properties changed? */
1335 if (!ofport_equal(&port->opp, &opp)) {
1336 ofport_modified(port, &opp);
1339 /* Install the newly opened netdev in case it has changed.
1340 * Don't close the old netdev yet in case port_modified has to
1341 * remove a retained reference to it.*/
1342 port->netdev = netdev;
1343 port->change_seq = netdev_change_seq(netdev);
1345 if (port->ofproto->ofproto_class->port_modified) {
1346 port->ofproto->ofproto_class->port_modified(port);
1349 netdev_close(old_netdev);
1351 /* If 'port' is nonnull then its name differs from 'name' and thus
1352 * we should delete it. If we think there's a port named 'name'
1353 * then its port number must be wrong now so delete it too. */
1355 ofport_remove(port);
1357 ofport_remove_with_name(ofproto, name);
1358 ofport_install(ofproto, netdev, &opp);
1361 /* Any port named 'name' is gone now. */
1362 ofport_remove_with_name(ofproto, name);
1364 ofproto_port_destroy(&ofproto_port);
1368 init_ports(struct ofproto *p)
1370 struct ofproto_port_dump dump;
1371 struct ofproto_port ofproto_port;
1373 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1374 uint16_t ofp_port = ofproto_port.ofp_port;
1375 if (ofproto_get_port(p, ofp_port)) {
1376 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1378 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1379 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1382 struct ofp_phy_port opp;
1383 struct netdev *netdev;
1385 netdev = ofport_open(&ofproto_port, &opp);
1387 ofport_install(p, netdev, &opp);
1396 ofproto_rule_destroy__(struct rule *rule)
1398 free(rule->actions);
1399 rule->ofproto->ofproto_class->rule_dealloc(rule);
1402 /* This function allows an ofproto implementation to destroy any rules that
1403 * remain when its ->destruct() function is called. The caller must have
1404 * already uninitialized any derived members of 'rule' (step 5 described in the
1405 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1406 * This function implements steps 6 and 7.
1408 * This function should only be called from an ofproto implementation's
1409 * ->destruct() function. It is not suitable elsewhere. */
1411 ofproto_rule_destroy(struct rule *rule)
1413 assert(!rule->pending);
1414 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1415 ofproto_rule_destroy__(rule);
1418 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1419 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1422 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1424 const union ofp_action *oa;
1427 if (out_port == OFPP_NONE) {
1430 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1431 if (action_outputs_to_port(oa, htons(out_port))) {
1438 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1439 * statistics appropriately. 'packet' must have at least sizeof(struct
1440 * ofp_packet_in) bytes of headroom.
1442 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1443 * with statistics for 'packet' either way.
1445 * Takes ownership of 'packet'. */
1447 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1451 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1453 flow_extract(packet, 0, in_port, &flow);
1454 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1457 /* Returns true if 'rule' should be hidden from the controller.
1459 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1460 * (e.g. by in-band control) and are intentionally hidden from the
1463 rule_is_hidden(const struct rule *rule)
1465 return rule->cr.priority > UINT16_MAX;
1469 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1471 ofconn_send_reply(ofconn, make_echo_reply(oh));
1476 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1478 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1479 struct ofp_switch_features *osf;
1481 struct ofport *port;
1485 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1486 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1488 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1489 osf->datapath_id = htonll(ofproto->datapath_id);
1490 osf->n_buffers = htonl(pktbuf_capacity());
1491 osf->n_tables = ofproto->n_tables;
1492 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1495 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1497 osf->actions = htonl(actions);
1499 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1500 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1503 ofconn_send_reply(ofconn, buf);
1508 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1510 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1512 struct ofp_switch_config *osc;
1516 /* Figure out flags. */
1517 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1518 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1521 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1522 osc->flags = htons(flags);
1523 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1524 ofconn_send_reply(ofconn, buf);
1530 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1532 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1533 uint16_t flags = ntohs(osc->flags);
1535 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1536 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1537 switch (flags & OFPC_FRAG_MASK) {
1538 case OFPC_FRAG_NORMAL:
1539 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1541 case OFPC_FRAG_DROP:
1542 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1545 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1551 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1556 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1557 * error message code (composed with ofp_mkerr()) for the caller to propagate
1558 * upward. Otherwise, returns 0.
1560 * The log message mentions 'msg_type'. */
1562 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1564 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1565 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1566 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1567 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1570 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1577 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1579 struct ofproto *p = ofconn_get_ofproto(ofconn);
1580 struct ofp_packet_out *opo;
1581 struct ofpbuf payload, *buffer;
1582 union ofp_action *ofp_actions;
1583 struct ofpbuf request;
1585 size_t n_ofp_actions;
1589 COVERAGE_INC(ofproto_packet_out);
1591 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1596 /* Get ofp_packet_out. */
1597 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1598 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1601 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1602 &ofp_actions, &n_ofp_actions);
1608 if (opo->buffer_id != htonl(UINT32_MAX)) {
1609 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1611 if (error || !buffer) {
1620 /* Send out packet. */
1621 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1622 error = p->ofproto_class->packet_out(p, &payload, &flow,
1623 ofp_actions, n_ofp_actions);
1624 ofpbuf_delete(buffer);
1630 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1632 ovs_be32 old_config = port->opp.config;
1634 mask &= config ^ port->opp.config;
1635 if (mask & htonl(OFPPC_PORT_DOWN)) {
1636 if (config & htonl(OFPPC_PORT_DOWN)) {
1637 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1639 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1643 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1644 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1645 OFPPC_NO_PACKET_IN));
1646 if (port->opp.config != old_config) {
1647 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1652 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1654 struct ofproto *p = ofconn_get_ofproto(ofconn);
1655 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1656 struct ofport *port;
1659 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1664 port = ofproto_get_port(p, ntohs(opm->port_no));
1666 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1667 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1668 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1670 update_port_config(port, opm->config, opm->mask);
1671 if (opm->advertise) {
1672 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1679 handle_desc_stats_request(struct ofconn *ofconn,
1680 const struct ofp_stats_msg *request)
1682 struct ofproto *p = ofconn_get_ofproto(ofconn);
1683 struct ofp_desc_stats *ods;
1686 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1687 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1688 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1689 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1690 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1691 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1692 ofconn_send_reply(ofconn, msg);
1698 handle_table_stats_request(struct ofconn *ofconn,
1699 const struct ofp_stats_msg *request)
1701 struct ofproto *p = ofconn_get_ofproto(ofconn);
1702 struct ofp_table_stats *ots;
1706 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1708 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1709 for (i = 0; i < p->n_tables; i++) {
1710 ots[i].table_id = i;
1711 sprintf(ots[i].name, "table%zu", i);
1712 ots[i].wildcards = htonl(OFPFW_ALL);
1713 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1714 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1717 p->ofproto_class->get_tables(p, ots);
1719 ofconn_send_reply(ofconn, msg);
1724 append_port_stat(struct ofport *port, struct list *replies)
1726 struct netdev_stats stats;
1727 struct ofp_port_stats *ops;
1729 /* Intentionally ignore return value, since errors will set
1730 * 'stats' to all-1s, which is correct for OpenFlow, and
1731 * netdev_get_stats() will log errors. */
1732 netdev_get_stats(port->netdev, &stats);
1734 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1735 ops->port_no = port->opp.port_no;
1736 memset(ops->pad, 0, sizeof ops->pad);
1737 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1738 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1739 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1740 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1741 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1742 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1743 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1744 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1745 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1746 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1747 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1748 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1752 handle_port_stats_request(struct ofconn *ofconn,
1753 const struct ofp_port_stats_request *psr)
1755 struct ofproto *p = ofconn_get_ofproto(ofconn);
1756 struct ofport *port;
1757 struct list replies;
1759 ofputil_start_stats_reply(&psr->osm, &replies);
1760 if (psr->port_no != htons(OFPP_NONE)) {
1761 port = ofproto_get_port(p, ntohs(psr->port_no));
1763 append_port_stat(port, &replies);
1766 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1767 append_port_stat(port, &replies);
1771 ofconn_send_replies(ofconn, &replies);
1776 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1778 long long int msecs = time_msec() - start;
1779 *sec = msecs / 1000;
1780 *nsec = (msecs % 1000) * (1000 * 1000);
1783 static struct classifier *
1784 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1786 if (table_id == 0xff) {
1787 return &ofproto->tables[0];
1788 } else if (table_id < ofproto->n_tables) {
1789 return &ofproto->tables[table_id];
1791 /* It would probably be better to reply with an error but there doesn't
1792 * seem to be any appropriate value, so that might just be
1794 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1800 static struct classifier *
1801 next_matching_table(struct ofproto *ofproto,
1802 struct classifier *cls, uint8_t table_id)
1804 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1809 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1812 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1815 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1816 * only once, for that table.
1818 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1819 * and does not enter the loop at all.
1821 * All parameters are evaluated multiple times.
1823 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1824 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1826 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1828 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1829 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1830 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1833 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1834 * to 'out_port' are included.
1836 * Hidden rules are always omitted.
1838 * Returns 0 on success, otherwise an OpenFlow error code. */
1840 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1841 const struct cls_rule *match, uint16_t out_port,
1844 struct classifier *cls;
1847 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1848 struct cls_cursor cursor;
1851 cls_cursor_init(&cursor, cls, match);
1852 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1853 if (rule->pending) {
1854 return OFPROTO_POSTPONE;
1856 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1857 list_push_back(rules, &rule->ofproto_node);
1864 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1865 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1866 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1869 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1870 * to 'out_port' are included.
1872 * Hidden rules are always omitted.
1874 * Returns 0 on success, otherwise an OpenFlow error code. */
1876 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1877 const struct cls_rule *match, uint16_t out_port,
1880 struct classifier *cls;
1883 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1886 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1888 if (rule->pending) {
1889 return OFPROTO_POSTPONE;
1891 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1892 list_push_back(rules, &rule->ofproto_node);
1900 handle_flow_stats_request(struct ofconn *ofconn,
1901 const struct ofp_stats_msg *osm)
1903 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1904 struct ofputil_flow_stats_request fsr;
1905 struct list replies;
1910 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1915 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1916 fsr.out_port, &rules);
1921 ofputil_start_stats_reply(osm, &replies);
1922 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1923 struct ofputil_flow_stats fs;
1926 fs.cookie = rule->flow_cookie;
1927 fs.table_id = rule->table_id;
1928 calc_flow_duration__(rule->created, &fs.duration_sec,
1930 fs.idle_timeout = rule->idle_timeout;
1931 fs.hard_timeout = rule->hard_timeout;
1932 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1934 fs.actions = rule->actions;
1935 fs.n_actions = rule->n_actions;
1936 ofputil_append_flow_stats_reply(&fs, &replies);
1938 ofconn_send_replies(ofconn, &replies);
1944 flow_stats_ds(struct rule *rule, struct ds *results)
1946 uint64_t packet_count, byte_count;
1948 rule->ofproto->ofproto_class->rule_get_stats(rule,
1949 &packet_count, &byte_count);
1951 if (rule->table_id != 0) {
1952 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1954 ds_put_format(results, "duration=%llds, ",
1955 (time_msec() - rule->created) / 1000);
1956 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1957 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1958 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1959 cls_rule_format(&rule->cr, results);
1960 ds_put_char(results, ',');
1961 if (rule->n_actions > 0) {
1962 ofp_print_actions(results, rule->actions, rule->n_actions);
1964 ds_put_cstr(results, "drop");
1966 ds_put_cstr(results, "\n");
1969 /* Adds a pretty-printed description of all flows to 'results', including
1970 * hidden flows (e.g., set up by in-band control). */
1972 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1974 struct classifier *cls;
1976 OFPROTO_FOR_EACH_TABLE (cls, p) {
1977 struct cls_cursor cursor;
1980 cls_cursor_init(&cursor, cls, NULL);
1981 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1982 flow_stats_ds(rule, results);
1987 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1988 * '*engine_type' and '*engine_id', respectively. */
1990 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1991 uint8_t *engine_type, uint8_t *engine_id)
1993 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1996 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1997 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1998 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
2000 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2002 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2003 return (ofport && ofproto->ofproto_class->get_cfm_fault
2004 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2008 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2009 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2010 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2011 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2013 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2014 uint16_t ofp_port, const uint64_t **rmps,
2017 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2021 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2022 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2028 handle_aggregate_stats_request(struct ofconn *ofconn,
2029 const struct ofp_stats_msg *osm)
2031 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2032 struct ofputil_flow_stats_request request;
2033 struct ofputil_aggregate_stats stats;
2034 bool unknown_packets, unknown_bytes;
2035 struct ofpbuf *reply;
2040 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2045 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2046 request.out_port, &rules);
2051 memset(&stats, 0, sizeof stats);
2052 unknown_packets = unknown_bytes = false;
2053 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2054 uint64_t packet_count;
2055 uint64_t byte_count;
2057 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2060 if (packet_count == UINT64_MAX) {
2061 unknown_packets = true;
2063 stats.packet_count += packet_count;
2066 if (byte_count == UINT64_MAX) {
2067 unknown_bytes = true;
2069 stats.byte_count += byte_count;
2074 if (unknown_packets) {
2075 stats.packet_count = UINT64_MAX;
2077 if (unknown_bytes) {
2078 stats.byte_count = UINT64_MAX;
2081 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2082 ofconn_send_reply(ofconn, reply);
2087 struct queue_stats_cbdata {
2088 struct ofport *ofport;
2089 struct list replies;
2093 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2094 const struct netdev_queue_stats *stats)
2096 struct ofp_queue_stats *reply;
2098 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2099 reply->port_no = cbdata->ofport->opp.port_no;
2100 memset(reply->pad, 0, sizeof reply->pad);
2101 reply->queue_id = htonl(queue_id);
2102 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2103 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2104 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2108 handle_queue_stats_dump_cb(uint32_t queue_id,
2109 struct netdev_queue_stats *stats,
2112 struct queue_stats_cbdata *cbdata = cbdata_;
2114 put_queue_stats(cbdata, queue_id, stats);
2118 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2119 struct queue_stats_cbdata *cbdata)
2121 cbdata->ofport = port;
2122 if (queue_id == OFPQ_ALL) {
2123 netdev_dump_queue_stats(port->netdev,
2124 handle_queue_stats_dump_cb, cbdata);
2126 struct netdev_queue_stats stats;
2128 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2129 put_queue_stats(cbdata, queue_id, &stats);
2135 handle_queue_stats_request(struct ofconn *ofconn,
2136 const struct ofp_queue_stats_request *qsr)
2138 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2139 struct queue_stats_cbdata cbdata;
2140 struct ofport *port;
2141 unsigned int port_no;
2144 COVERAGE_INC(ofproto_queue_req);
2146 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2148 port_no = ntohs(qsr->port_no);
2149 queue_id = ntohl(qsr->queue_id);
2150 if (port_no == OFPP_ALL) {
2151 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2152 handle_queue_stats_for_port(port, queue_id, &cbdata);
2154 } else if (port_no < OFPP_MAX) {
2155 port = ofproto_get_port(ofproto, port_no);
2157 handle_queue_stats_for_port(port, queue_id, &cbdata);
2160 ofpbuf_list_delete(&cbdata.replies);
2161 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2163 ofconn_send_replies(ofconn, &cbdata.replies);
2169 is_flow_deletion_pending(const struct ofproto *ofproto,
2170 const struct cls_rule *cls_rule,
2173 if (!hmap_is_empty(&ofproto->deletions)) {
2174 struct ofoperation *op;
2176 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2177 cls_rule_hash(cls_rule, table_id),
2178 &ofproto->deletions) {
2179 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2188 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2189 * in which no matching flow already exists in the flow table.
2191 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2192 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2193 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2195 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2198 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2199 struct ofputil_flow_mod *fm, const struct ofp_header *request)
2201 struct classifier *table;
2202 struct ofopgroup *group;
2203 struct rule *victim;
2207 /* Check for overlap, if requested. */
2208 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2209 struct classifier *cls;
2211 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2212 if (classifier_rule_overlaps(cls, &fm->cr)) {
2213 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2219 if (fm->table_id == 0xff) {
2221 if (ofproto->ofproto_class->rule_choose_table) {
2222 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2227 assert(table_id < ofproto->n_tables);
2228 table = &ofproto->tables[table_id];
2230 table = &ofproto->tables[0];
2232 } else if (fm->table_id < ofproto->n_tables) {
2233 table = &ofproto->tables[fm->table_id];
2235 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2238 /* Serialize against pending deletion. */
2239 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2240 return OFPROTO_POSTPONE;
2243 /* Allocate new rule. */
2244 rule = ofproto->ofproto_class->rule_alloc();
2246 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2247 ofproto->name, strerror(error));
2250 rule->ofproto = ofproto;
2252 rule->pending = NULL;
2253 rule->flow_cookie = fm->cookie;
2254 rule->created = time_msec();
2255 rule->idle_timeout = fm->idle_timeout;
2256 rule->hard_timeout = fm->hard_timeout;
2257 rule->table_id = table - ofproto->tables;
2258 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2259 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2260 rule->n_actions = fm->n_actions;
2262 /* Insert new rule. */
2263 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2264 if (victim && victim->pending) {
2265 error = OFPROTO_POSTPONE;
2268 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2269 : ofopgroup_create(ofproto));
2270 ofoperation_create(group, rule, OFOPERATION_ADD);
2271 rule->pending->victim = victim;
2273 error = ofproto->ofproto_class->rule_construct(rule);
2275 ofoperation_destroy(rule->pending);
2277 ofopgroup_submit(group);
2280 /* Back out if an error occurred. */
2283 classifier_replace(table, &victim->cr);
2285 classifier_remove(table, &rule->cr);
2287 ofproto_rule_destroy__(rule);
2292 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2294 /* Modifies the rules listed in 'rules', changing their actions to match those
2297 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2300 * Returns 0 on success, otherwise an OpenFlow error code. */
2302 modify_flows__(struct ofconn *ofconn, const struct ofputil_flow_mod *fm,
2303 const struct ofp_header *request, struct list *rules)
2305 struct ofopgroup *group;
2308 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2309 LIST_FOR_EACH (rule, ofproto_node, rules) {
2310 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2311 rule->actions, rule->n_actions)) {
2312 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2313 rule->pending->actions = rule->actions;
2314 rule->pending->n_actions = rule->n_actions;
2315 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2316 rule->n_actions = fm->n_actions;
2317 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2319 rule->flow_cookie = fm->cookie;
2321 ofopgroup_submit(group);
2326 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2327 * encoded by ofp_mkerr() on failure.
2329 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2332 modify_flows_loose(struct ofconn *ofconn, struct ofputil_flow_mod *fm,
2333 const struct ofp_header *request)
2335 struct ofproto *p = ofconn_get_ofproto(ofconn);
2339 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2340 return (error ? error
2341 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2342 : modify_flows__(ofconn, fm, request, &rules));
2345 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2346 * code as encoded by ofp_mkerr() on failure.
2348 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2351 modify_flow_strict(struct ofconn *ofconn, struct ofputil_flow_mod *fm,
2352 const struct ofp_header *request)
2354 struct ofproto *p = ofconn_get_ofproto(ofconn);
2358 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2359 return (error ? error
2360 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2361 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2366 /* OFPFC_DELETE implementation. */
2368 /* Deletes the rules listed in 'rules'.
2370 * Returns 0 on success, otherwise an OpenFlow error code. */
2372 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2375 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2376 struct rule *rule, *next;
2377 struct ofopgroup *group;
2379 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2380 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2381 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2383 ofoperation_create(group, rule, OFOPERATION_DELETE);
2384 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2385 rule->ofproto->ofproto_class->rule_destruct(rule);
2387 ofopgroup_submit(group);
2392 /* Implements OFPFC_DELETE. */
2394 delete_flows_loose(struct ofconn *ofconn, const struct ofputil_flow_mod *fm,
2395 const struct ofp_header *request)
2397 struct ofproto *p = ofconn_get_ofproto(ofconn);
2401 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2403 return (error ? error
2404 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2408 /* Implements OFPFC_DELETE_STRICT. */
2410 delete_flow_strict(struct ofconn *ofconn, struct ofputil_flow_mod *fm,
2411 const struct ofp_header *request)
2413 struct ofproto *p = ofconn_get_ofproto(ofconn);
2417 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2419 return (error ? error
2420 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2426 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2428 struct ofputil_flow_removed fr;
2430 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2435 fr.cookie = rule->flow_cookie;
2437 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2438 fr.idle_timeout = rule->idle_timeout;
2439 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2442 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2445 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2446 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2449 * ofproto implementation ->run() functions should use this function to expire
2450 * OpenFlow flows. */
2452 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2454 struct ofproto *ofproto = rule->ofproto;
2455 struct ofopgroup *group;
2457 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2459 ofproto_rule_send_removed(rule, reason);
2461 group = ofopgroup_create(ofproto);
2462 ofoperation_create(group, rule, OFOPERATION_DELETE);
2463 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2464 rule->ofproto->ofproto_class->rule_destruct(rule);
2465 ofopgroup_submit(group);
2469 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2471 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2472 struct ofputil_flow_mod fm;
2475 error = reject_slave_controller(ofconn, "flow_mod");
2480 if (ofproto->n_pending >= 50) {
2481 assert(!list_is_empty(&ofproto->pending));
2482 return OFPROTO_POSTPONE;
2485 error = ofputil_decode_flow_mod(&fm, oh,
2486 ofconn_get_flow_mod_table_id(ofconn));
2491 /* We do not support the emergency flow cache. It will hopefully get
2492 * dropped from OpenFlow in the near future. */
2493 if (fm.flags & OFPFF_EMERG) {
2494 /* There isn't a good fit for an error code, so just state that the
2495 * flow table is full. */
2496 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2499 switch (fm.command) {
2501 return add_flow(ofproto, ofconn, &fm, oh);
2504 return modify_flows_loose(ofconn, &fm, oh);
2506 case OFPFC_MODIFY_STRICT:
2507 return modify_flow_strict(ofconn, &fm, oh);
2510 return delete_flows_loose(ofconn, &fm, oh);
2512 case OFPFC_DELETE_STRICT:
2513 return delete_flow_strict(ofconn, &fm, oh);
2516 if (fm.command > 0xff) {
2517 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2518 "flow_mod_table_id extension is not enabled");
2520 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2525 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2527 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2528 struct nx_role_request *reply;
2532 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2533 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2534 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2537 role = ntohl(nrr->role);
2538 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2539 && role != NX_ROLE_SLAVE) {
2540 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2542 /* There's no good error code for this. */
2543 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2546 if (ofconn_get_role(ofconn) != role
2547 && ofconn_has_pending_opgroups(ofconn)) {
2548 return OFPROTO_POSTPONE;
2551 ofconn_set_role(ofconn, role);
2553 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2554 reply->role = htonl(role);
2555 ofconn_send_reply(ofconn, buf);
2561 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2562 const struct ofp_header *oh)
2564 const struct nxt_flow_mod_table_id *msg
2565 = (const struct nxt_flow_mod_table_id *) oh;
2567 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2572 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2574 const struct nxt_set_flow_format *msg
2575 = (const struct nxt_set_flow_format *) oh;
2578 format = ntohl(msg->format);
2579 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2580 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2583 if (format != ofconn_get_flow_format(ofconn)
2584 && ofconn_has_pending_opgroups(ofconn)) {
2585 /* Avoid sending async messages in surprising flow format. */
2586 return OFPROTO_POSTPONE;
2589 ofconn_set_flow_format(ofconn, format);
2594 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2596 struct ofp_header *ob;
2599 if (ofconn_has_pending_opgroups(ofconn)) {
2600 return OFPROTO_POSTPONE;
2603 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2604 ofconn_send_reply(ofconn, buf);
2609 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2611 const struct ofp_header *oh = msg->data;
2612 const struct ofputil_msg_type *type;
2615 error = ofputil_decode_msg_type(oh, &type);
2620 switch (ofputil_msg_type_code(type)) {
2621 /* OpenFlow requests. */
2622 case OFPUTIL_OFPT_ECHO_REQUEST:
2623 return handle_echo_request(ofconn, oh);
2625 case OFPUTIL_OFPT_FEATURES_REQUEST:
2626 return handle_features_request(ofconn, oh);
2628 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2629 return handle_get_config_request(ofconn, oh);
2631 case OFPUTIL_OFPT_SET_CONFIG:
2632 return handle_set_config(ofconn, msg->data);
2634 case OFPUTIL_OFPT_PACKET_OUT:
2635 return handle_packet_out(ofconn, oh);
2637 case OFPUTIL_OFPT_PORT_MOD:
2638 return handle_port_mod(ofconn, oh);
2640 case OFPUTIL_OFPT_FLOW_MOD:
2641 return handle_flow_mod(ofconn, oh);
2643 case OFPUTIL_OFPT_BARRIER_REQUEST:
2644 return handle_barrier_request(ofconn, oh);
2646 /* OpenFlow replies. */
2647 case OFPUTIL_OFPT_ECHO_REPLY:
2650 /* Nicira extension requests. */
2651 case OFPUTIL_NXT_ROLE_REQUEST:
2652 return handle_role_request(ofconn, oh);
2654 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2655 return handle_nxt_flow_mod_table_id(ofconn, oh);
2657 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2658 return handle_nxt_set_flow_format(ofconn, oh);
2660 case OFPUTIL_NXT_FLOW_MOD:
2661 return handle_flow_mod(ofconn, oh);
2663 /* Statistics requests. */
2664 case OFPUTIL_OFPST_DESC_REQUEST:
2665 return handle_desc_stats_request(ofconn, msg->data);
2667 case OFPUTIL_OFPST_FLOW_REQUEST:
2668 case OFPUTIL_NXST_FLOW_REQUEST:
2669 return handle_flow_stats_request(ofconn, msg->data);
2671 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2672 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2673 return handle_aggregate_stats_request(ofconn, msg->data);
2675 case OFPUTIL_OFPST_TABLE_REQUEST:
2676 return handle_table_stats_request(ofconn, msg->data);
2678 case OFPUTIL_OFPST_PORT_REQUEST:
2679 return handle_port_stats_request(ofconn, msg->data);
2681 case OFPUTIL_OFPST_QUEUE_REQUEST:
2682 return handle_queue_stats_request(ofconn, msg->data);
2684 case OFPUTIL_MSG_INVALID:
2685 case OFPUTIL_OFPT_HELLO:
2686 case OFPUTIL_OFPT_ERROR:
2687 case OFPUTIL_OFPT_FEATURES_REPLY:
2688 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2689 case OFPUTIL_OFPT_PACKET_IN:
2690 case OFPUTIL_OFPT_FLOW_REMOVED:
2691 case OFPUTIL_OFPT_PORT_STATUS:
2692 case OFPUTIL_OFPT_BARRIER_REPLY:
2693 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2694 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2695 case OFPUTIL_OFPST_DESC_REPLY:
2696 case OFPUTIL_OFPST_FLOW_REPLY:
2697 case OFPUTIL_OFPST_QUEUE_REPLY:
2698 case OFPUTIL_OFPST_PORT_REPLY:
2699 case OFPUTIL_OFPST_TABLE_REPLY:
2700 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2701 case OFPUTIL_NXT_ROLE_REPLY:
2702 case OFPUTIL_NXT_FLOW_REMOVED:
2703 case OFPUTIL_NXST_FLOW_REPLY:
2704 case OFPUTIL_NXST_AGGREGATE_REPLY:
2706 if (VLOG_IS_WARN_ENABLED()) {
2707 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2708 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2711 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2712 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2714 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2720 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2722 int error = handle_openflow__(ofconn, ofp_msg);
2723 if (error && error != OFPROTO_POSTPONE) {
2724 ofconn_send_error(ofconn, ofp_msg->data, error);
2726 COVERAGE_INC(ofproto_recv_openflow);
2727 return error != OFPROTO_POSTPONE;
2730 /* Asynchronous operations. */
2732 /* Creates and returns a new ofopgroup that is not associated with any
2733 * OpenFlow connection.
2735 * The caller should add operations to the returned group with
2736 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2737 static struct ofopgroup *
2738 ofopgroup_create(struct ofproto *ofproto)
2740 struct ofopgroup *group = xzalloc(sizeof *group);
2741 group->ofproto = ofproto;
2742 list_init(&group->ofproto_node);
2743 list_init(&group->ops);
2744 list_init(&group->ofconn_node);
2748 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2749 * the ofopgroup eventually fails, then the error reply will include 'request'.
2750 * If the ofopgroup eventually succeeds, then the packet with buffer id
2751 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2753 * The caller should add operations to the returned group with
2754 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2755 static struct ofopgroup *
2756 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2757 const struct ofp_header *request,
2760 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2761 size_t request_len = ntohs(request->length);
2763 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2764 group->ofconn = ofconn;
2765 group->request = xmemdup(request, MIN(request_len, 64));
2766 group->buffer_id = buffer_id;
2771 /* Submits 'group' for processing.
2773 * If 'group' contains no operations (e.g. none were ever added, or all of the
2774 * ones that were added completed synchronously), then it is destroyed
2775 * immediately. Otherwise it is added to the ofproto's list of pending
2778 ofopgroup_submit(struct ofopgroup *group)
2780 if (list_is_empty(&group->ops)) {
2781 ofopgroup_destroy(group);
2783 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2784 group->ofproto->n_pending++;
2789 ofopgroup_destroy(struct ofopgroup *group)
2791 assert(list_is_empty(&group->ops));
2792 if (!list_is_empty(&group->ofproto_node)) {
2793 assert(group->ofproto->n_pending > 0);
2794 group->ofproto->n_pending--;
2795 list_remove(&group->ofproto_node);
2797 if (!list_is_empty(&group->ofconn_node)) {
2798 list_remove(&group->ofconn_node);
2800 ofconn_send_error(group->ofconn, group->request, group->error);
2802 connmgr_retry(group->ofproto->connmgr);
2804 free(group->request);
2808 /* Initiates a new operation on 'rule', of the specified 'type', within
2809 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2811 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2812 enum ofoperation_type type)
2814 struct ofoperation *op;
2816 assert(!rule->pending);
2818 op = rule->pending = xzalloc(sizeof *op);
2820 list_push_back(&group->ops, &op->group_node);
2824 op->flow_cookie = rule->flow_cookie;
2826 if (type == OFOPERATION_DELETE) {
2827 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2828 cls_rule_hash(&rule->cr, rule->table_id));
2833 ofoperation_destroy(struct ofoperation *op)
2835 struct ofopgroup *group = op->group;
2838 op->rule->pending = NULL;
2840 if (op->type == OFOPERATION_DELETE) {
2841 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2843 list_remove(&op->group_node);
2847 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2848 ofopgroup_destroy(group);
2852 /* Indicates that 'op' completed with status 'error', which is either 0 to
2853 * indicate success or an OpenFlow error code (constructed with
2854 * e.g. ofp_mkerr()).
2856 * If 'error' is 0, indicating success, the operation will be committed
2857 * permanently to the flow table. There is one interesting subcase:
2859 * - If 'op' is an "add flow" operation that is replacing an existing rule in
2860 * the flow table (the "victim" rule) by a new one, then the caller must
2861 * have uninitialized any derived state in the victim rule, as in step 5 in
2862 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
2863 * performs steps 6 and 7 for the victim rule, most notably by calling its
2864 * ->rule_dealloc() function.
2866 * If 'error' is nonzero, then generally the operation will be rolled back:
2868 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
2869 * restores the original rule. The caller must have uninitialized any
2870 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
2871 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
2872 * and 7 for the new rule, calling its ->rule_dealloc() function.
2874 * - If 'op' is a "modify flow" operation, ofproto restores the original
2877 * - 'op' must not be a "delete flow" operation. Removing a rule is not
2878 * allowed to fail. It must always succeed.
2880 * Please see the large comment in ofproto/ofproto-provider.h titled
2881 * "Asynchronous Operation Support" for more information. */
2883 ofoperation_complete(struct ofoperation *op, int error)
2885 struct ofopgroup *group = op->group;
2886 struct rule *rule = op->rule;
2887 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2889 assert(rule->pending == op);
2890 assert(op->status < 0);
2895 && op->type != OFOPERATION_DELETE
2897 && group->buffer_id != UINT32_MAX
2898 && list_is_singleton(&op->group_node)) {
2899 struct ofpbuf *packet;
2902 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2906 error = rule_execute(rule, in_port, packet);
2909 if (!group->error) {
2910 group->error = error;
2914 case OFOPERATION_ADD:
2917 ofproto_rule_destroy__(op->victim);
2921 classifier_replace(table, &op->victim->cr);
2924 classifier_remove(table, &rule->cr);
2926 ofproto_rule_destroy__(rule);
2931 case OFOPERATION_DELETE:
2933 ofproto_rule_destroy__(rule);
2937 case OFOPERATION_MODIFY:
2939 free(rule->actions);
2940 rule->actions = op->actions;
2941 rule->n_actions = op->n_actions;
2949 ofoperation_destroy(op);
2953 ofoperation_get_victim(struct ofoperation *op)
2955 assert(op->type == OFOPERATION_ADD);
2960 pick_datapath_id(const struct ofproto *ofproto)
2962 const struct ofport *port;
2964 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2966 uint8_t ea[ETH_ADDR_LEN];
2969 error = netdev_get_etheraddr(port->netdev, ea);
2971 return eth_addr_to_uint64(ea);
2973 VLOG_WARN("could not get MAC address for %s (%s)",
2974 netdev_get_name(port->netdev), strerror(error));
2976 return ofproto->fallback_dpid;
2980 pick_fallback_dpid(void)
2982 uint8_t ea[ETH_ADDR_LEN];
2983 eth_addr_nicira_random(ea);
2984 return eth_addr_to_uint64(ea);
2987 /* unixctl commands. */
2990 ofproto_lookup(const char *name)
2992 struct ofproto *ofproto;
2994 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2996 if (!strcmp(ofproto->name, name)) {
3004 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
3005 void *aux OVS_UNUSED)
3007 struct ofproto *ofproto;
3011 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
3012 ds_put_format(&results, "%s\n", ofproto->name);
3014 unixctl_command_reply(conn, 200, ds_cstr(&results));
3015 ds_destroy(&results);
3019 ofproto_unixctl_init(void)
3021 static bool registered;
3027 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);