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 "openflow/nicira-ext.h"
37 #include "openflow/openflow.h"
41 #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->fallback_dpid = pick_fallback_dpid();
324 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
325 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
326 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
327 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
328 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
329 hmap_init(&ofproto->ports);
330 shash_init(&ofproto->port_by_name);
331 ofproto->tables = NULL;
332 ofproto->n_tables = 0;
333 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
334 ofproto->state = S_OPENFLOW;
335 list_init(&ofproto->pending);
336 hmap_init(&ofproto->deletions);
338 error = ofproto->ofproto_class->construct(ofproto);
340 VLOG_ERR("failed to open datapath %s: %s",
341 datapath_name, strerror(error));
342 ofproto_destroy__(ofproto);
345 assert(ofproto->n_tables > 0);
347 ofproto->datapath_id = pick_datapath_id(ofproto);
348 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
356 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
358 uint64_t old_dpid = p->datapath_id;
359 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
360 if (p->datapath_id != old_dpid) {
361 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
363 /* Force all active connections to reconnect, since there is no way to
364 * notify a controller that the datapath ID has changed. */
365 ofproto_reconnect_controllers(p);
370 ofproto_set_controllers(struct ofproto *p,
371 const struct ofproto_controller *controllers,
372 size_t n_controllers)
374 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
378 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
380 connmgr_set_fail_mode(p->connmgr, fail_mode);
383 /* Drops the connections between 'ofproto' and all of its controllers, forcing
384 * them to reconnect. */
386 ofproto_reconnect_controllers(struct ofproto *ofproto)
388 connmgr_reconnect(ofproto->connmgr);
391 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
392 * in-band control should guarantee access, in the same way that in-band
393 * control guarantees access to OpenFlow controllers. */
395 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
396 const struct sockaddr_in *extras, size_t n)
398 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
401 /* Sets the OpenFlow queue used by flows set up by in-band control on
402 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
403 * flows will use the default queue. */
405 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
407 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
411 ofproto_set_desc(struct ofproto *p,
412 const char *mfr_desc, const char *hw_desc,
413 const char *sw_desc, const char *serial_desc,
416 struct ofp_desc_stats *ods;
419 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
420 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
421 sizeof ods->mfr_desc);
424 p->mfr_desc = xstrdup(mfr_desc);
427 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
428 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
429 sizeof ods->hw_desc);
432 p->hw_desc = xstrdup(hw_desc);
435 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
436 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
437 sizeof ods->sw_desc);
440 p->sw_desc = xstrdup(sw_desc);
443 if (strlen(serial_desc) >= sizeof ods->serial_num) {
444 VLOG_WARN("truncating serial_desc, must be less than %zu "
446 sizeof ods->serial_num);
448 free(p->serial_desc);
449 p->serial_desc = xstrdup(serial_desc);
452 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
453 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
454 sizeof ods->dp_desc);
457 p->dp_desc = xstrdup(dp_desc);
462 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
464 return connmgr_set_snoops(ofproto->connmgr, snoops);
468 ofproto_set_netflow(struct ofproto *ofproto,
469 const struct netflow_options *nf_options)
471 if (nf_options && sset_is_empty(&nf_options->collectors)) {
475 if (ofproto->ofproto_class->set_netflow) {
476 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
478 return nf_options ? EOPNOTSUPP : 0;
483 ofproto_set_sflow(struct ofproto *ofproto,
484 const struct ofproto_sflow_options *oso)
486 if (oso && sset_is_empty(&oso->targets)) {
490 if (ofproto->ofproto_class->set_sflow) {
491 return ofproto->ofproto_class->set_sflow(ofproto, oso);
493 return oso ? EOPNOTSUPP : 0;
497 /* Connectivity Fault Management configuration. */
499 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
501 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
503 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
504 if (ofport && ofproto->ofproto_class->set_cfm) {
505 ofproto->ofproto_class->set_cfm(ofport, NULL);
509 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
510 * basic configuration from the configuration members in 'cfm', and the remote
511 * maintenance point ID from remote_mpid. Ignores the statistics members of
514 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
516 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
517 const struct cfm_settings *s)
519 struct ofport *ofport;
522 ofport = ofproto_get_port(ofproto, ofp_port);
524 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
525 ofproto->name, ofp_port);
529 /* XXX: For configuration simplicity, we only support one remote_mpid
530 * outside of the CFM module. It's not clear if this is the correct long
531 * term solution or not. */
532 error = (ofproto->ofproto_class->set_cfm
533 ? ofproto->ofproto_class->set_cfm(ofport, s)
536 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
537 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
542 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
543 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
544 * 0 if LACP partner information is not current (generally indicating a
545 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
547 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
549 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
550 return (ofport && ofproto->ofproto_class->port_is_lacp_current
551 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
557 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
558 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
559 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
560 * configuration plus, if there is more than one slave, a bonding
563 * If 'aux' is already registered then this function updates its configuration
564 * to 's'. Otherwise, this function registers a new bundle.
566 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
569 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
570 const struct ofproto_bundle_settings *s)
572 return (ofproto->ofproto_class->bundle_set
573 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
577 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
578 * If no such bundle has been registered, this has no effect. */
580 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
582 return ofproto_bundle_register(ofproto, aux, NULL);
586 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
587 * If 'aux' is already registered then this function updates its configuration
588 * to 's'. Otherwise, this function registers a new mirror.
590 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
593 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
594 const struct ofproto_mirror_settings *s)
596 return (ofproto->ofproto_class->mirror_set
597 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
601 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
602 * If no mirror has been registered, this has no effect. */
604 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
606 return ofproto_mirror_register(ofproto, aux, NULL);
609 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
610 * which all packets are flooded, instead of using MAC learning. If
611 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
613 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
616 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
618 return (ofproto->ofproto_class->set_flood_vlans
619 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
623 /* Returns true if 'aux' is a registered bundle that is currently in use as the
624 * output for a mirror. */
626 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
628 return (ofproto->ofproto_class->is_mirror_output_bundle
629 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
634 ofproto_has_snoops(const struct ofproto *ofproto)
636 return connmgr_has_snoops(ofproto->connmgr);
640 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
642 connmgr_get_snoops(ofproto->connmgr, snoops);
646 ofproto_flush__(struct ofproto *ofproto)
648 struct classifier *table;
649 struct ofopgroup *group;
651 if (ofproto->ofproto_class->flush) {
652 ofproto->ofproto_class->flush(ofproto);
655 group = ofopgroup_create(ofproto);
656 for (table = ofproto->tables; table < &ofproto->tables[ofproto->n_tables];
658 struct rule *rule, *next_rule;
659 struct cls_cursor cursor;
661 cls_cursor_init(&cursor, table, NULL);
662 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
663 if (!rule->pending) {
664 ofoperation_create(group, rule, OFOPERATION_DELETE);
665 classifier_remove(table, &rule->cr);
666 ofproto->ofproto_class->rule_destruct(rule);
670 ofopgroup_submit(group);
674 ofproto_destroy__(struct ofproto *ofproto)
678 assert(list_is_empty(&ofproto->pending));
680 connmgr_destroy(ofproto->connmgr);
682 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
685 free(ofproto->mfr_desc);
686 free(ofproto->hw_desc);
687 free(ofproto->sw_desc);
688 free(ofproto->serial_desc);
689 free(ofproto->dp_desc);
690 hmap_destroy(&ofproto->ports);
691 shash_destroy(&ofproto->port_by_name);
693 for (i = 0; i < ofproto->n_tables; i++) {
694 assert(classifier_is_empty(&ofproto->tables[i]));
695 classifier_destroy(&ofproto->tables[i]);
697 free(ofproto->tables);
699 hmap_destroy(&ofproto->deletions);
701 ofproto->ofproto_class->dealloc(ofproto);
705 ofproto_destroy(struct ofproto *p)
707 struct ofport *ofport, *next_ofport;
714 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
715 ofport_destroy(ofport);
718 p->ofproto_class->destruct(p);
719 ofproto_destroy__(p);
722 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
723 * kernel datapath, for example, this destroys the datapath in the kernel, and
724 * with the netdev-based datapath, it tears down the data structures that
725 * represent the datapath.
727 * The datapath should not be currently open as an ofproto. */
729 ofproto_delete(const char *name, const char *type)
731 const struct ofproto_class *class = ofproto_class_find__(type);
732 return (!class ? EAFNOSUPPORT
733 : !class->del ? EACCES
734 : class->del(type, name));
738 process_port_change(struct ofproto *ofproto, int error, char *devname)
740 if (error == ENOBUFS) {
741 reinit_ports(ofproto);
743 update_port(ofproto, devname);
749 ofproto_run(struct ofproto *p)
751 struct ofport *ofport;
755 error = p->ofproto_class->run(p);
756 if (error == ENODEV) {
757 /* Someone destroyed the datapath behind our back. The caller
758 * better destroy us and give up, because we're just going to
759 * spin from here on out. */
760 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
761 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
766 if (p->ofproto_class->port_poll) {
767 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
768 process_port_change(p, error, devname);
772 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
773 unsigned int change_seq = netdev_change_seq(ofport->netdev);
774 if (ofport->change_seq != change_seq) {
775 ofport->change_seq = change_seq;
776 update_port(p, netdev_get_name(ofport->netdev));
783 connmgr_run(p->connmgr, handle_openflow);
787 connmgr_run(p->connmgr, NULL);
789 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
790 connmgr_flushed(p->connmgr);
791 p->state = S_OPENFLOW;
803 ofproto_wait(struct ofproto *p)
805 struct ofport *ofport;
807 p->ofproto_class->wait(p);
808 if (p->ofproto_class->port_poll_wait) {
809 p->ofproto_class->port_poll_wait(p);
812 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
813 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
814 poll_immediate_wake();
820 connmgr_wait(p->connmgr, true);
824 connmgr_wait(p->connmgr, false);
825 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
826 poll_immediate_wake();
833 ofproto_is_alive(const struct ofproto *p)
835 return connmgr_has_controllers(p->connmgr);
839 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
842 connmgr_get_controller_info(ofproto->connmgr, info);
846 ofproto_free_ofproto_controller_info(struct shash *info)
848 connmgr_free_controller_info(info);
851 /* Makes a deep copy of 'old' into 'port'. */
853 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
855 port->name = xstrdup(old->name);
856 port->type = xstrdup(old->type);
857 port->ofp_port = old->ofp_port;
860 /* Frees memory allocated to members of 'ofproto_port'.
862 * Do not call this function on an ofproto_port obtained from
863 * ofproto_port_dump_next(): that function retains ownership of the data in the
866 ofproto_port_destroy(struct ofproto_port *ofproto_port)
868 free(ofproto_port->name);
869 free(ofproto_port->type);
872 /* Initializes 'dump' to begin dumping the ports in an ofproto.
874 * This function provides no status indication. An error status for the entire
875 * dump operation is provided when it is completed by calling
876 * ofproto_port_dump_done().
879 ofproto_port_dump_start(struct ofproto_port_dump *dump,
880 const struct ofproto *ofproto)
882 dump->ofproto = ofproto;
883 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
887 /* Attempts to retrieve another port from 'dump', which must have been created
888 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
889 * 'port' and returns true. On failure, returns false.
891 * Failure might indicate an actual error or merely that the last port has been
892 * dumped. An error status for the entire dump operation is provided when it
893 * is completed by calling ofproto_port_dump_done().
895 * The ofproto owns the data stored in 'port'. It will remain valid until at
896 * least the next time 'dump' is passed to ofproto_port_dump_next() or
897 * ofproto_port_dump_done(). */
899 ofproto_port_dump_next(struct ofproto_port_dump *dump,
900 struct ofproto_port *port)
902 const struct ofproto *ofproto = dump->ofproto;
908 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
911 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
917 /* Completes port table dump operation 'dump', which must have been created
918 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
919 * error-free, otherwise a positive errno value describing the problem. */
921 ofproto_port_dump_done(struct ofproto_port_dump *dump)
923 const struct ofproto *ofproto = dump->ofproto;
925 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
928 return dump->error == EOF ? 0 : dump->error;
931 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
932 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
933 * is non-null). On failure, returns a positive errno value and sets
934 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
936 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
942 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
944 update_port(ofproto, netdev_get_name(netdev));
947 *ofp_portp = error ? OFPP_NONE : ofp_port;
952 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
953 * initializes '*port' appropriately; on failure, returns a positive errno
956 * The caller owns the data in 'ofproto_port' and must free it with
957 * ofproto_port_destroy() when it is no longer needed. */
959 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
960 struct ofproto_port *port)
964 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
966 memset(port, 0, sizeof *port);
971 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
972 * Returns 0 if successful, otherwise a positive errno. */
974 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
976 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
977 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
980 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
981 if (!error && ofport) {
982 /* 'name' is the netdev's name and update_port() is going to close the
983 * netdev. Just in case update_port() refers to 'name' after it
984 * destroys 'ofport', make a copy of it around the update_port()
986 char *devname = xstrdup(name);
987 update_port(ofproto, devname);
993 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
994 * performs the 'n_actions' actions in 'actions'. The new flow will not
997 * If cls_rule->priority is in the range of priorities supported by OpenFlow
998 * (0...65535, inclusive) then the flow will be visible to OpenFlow
999 * controllers; otherwise, it will be hidden.
1001 * The caller retains ownership of 'cls_rule' and 'actions'.
1003 * This is a helper function for in-band control and fail-open. */
1005 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1006 const union ofp_action *actions, size_t n_actions)
1008 const struct rule *rule;
1010 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1011 &ofproto->tables[0], cls_rule));
1012 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1013 actions, n_actions)) {
1016 memset(&fm, 0, sizeof fm);
1018 fm.buffer_id = UINT32_MAX;
1019 fm.actions = (union ofp_action *) actions;
1020 fm.n_actions = n_actions;
1021 add_flow(ofproto, NULL, &fm, NULL);
1025 /* Searches for a rule with matching criteria exactly equal to 'target' in
1026 * ofproto's table 0 and, if it finds one, deletes it.
1028 * This is a helper function for in-band control and fail-open. */
1030 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1034 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1035 &ofproto->tables[0], target));
1037 /* No such rule -> success. */
1039 } else if (rule->pending) {
1040 /* An operation on the rule is already pending -> failure.
1041 * Caller must retry later if it's important. */
1044 /* Initiate deletion -> success. */
1045 struct ofopgroup *group = ofopgroup_create(ofproto);
1046 ofoperation_create(group, rule, OFOPERATION_DELETE);
1047 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1048 rule->ofproto->ofproto_class->rule_destruct(rule);
1049 ofopgroup_submit(group);
1055 /* Starts the process of deleting all of the flows from all of ofproto's flow
1056 * tables and then reintroducing the flows required by in-band control and
1057 * fail-open. The process will complete in a later call to ofproto_run(). */
1059 ofproto_flush_flows(struct ofproto *ofproto)
1061 COVERAGE_INC(ofproto_flush);
1062 ofproto->state = S_FLUSH;
1066 reinit_ports(struct ofproto *p)
1068 struct ofproto_port_dump dump;
1069 struct sset devnames;
1070 struct ofport *ofport;
1071 struct ofproto_port ofproto_port;
1072 const char *devname;
1074 COVERAGE_INC(ofproto_reinit_ports);
1076 sset_init(&devnames);
1077 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1078 sset_add(&devnames, netdev_get_name(ofport->netdev));
1080 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1081 sset_add(&devnames, ofproto_port.name);
1084 SSET_FOR_EACH (devname, &devnames) {
1085 update_port(p, devname);
1087 sset_destroy(&devnames);
1090 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1091 * netdev cannot be opened. On success, also fills in 'opp'. */
1092 static struct netdev *
1093 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1095 uint32_t curr, advertised, supported, peer;
1096 struct netdev_options netdev_options;
1097 enum netdev_flags flags;
1098 struct netdev *netdev;
1101 memset(&netdev_options, 0, sizeof netdev_options);
1102 netdev_options.name = ofproto_port->name;
1103 netdev_options.type = ofproto_port->type;
1104 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1106 error = netdev_open(&netdev_options, &netdev);
1108 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1109 "cannot be opened (%s)",
1110 ofproto_port->name, ofproto_port->ofp_port,
1111 ofproto_port->name, strerror(error));
1115 netdev_get_flags(netdev, &flags);
1116 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1118 opp->port_no = htons(ofproto_port->ofp_port);
1119 netdev_get_etheraddr(netdev, opp->hw_addr);
1120 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1121 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1122 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1123 opp->curr = htonl(curr);
1124 opp->advertised = htonl(advertised);
1125 opp->supported = htonl(supported);
1126 opp->peer = htonl(peer);
1131 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1132 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1135 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1137 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1138 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1139 && a->state == b->state
1140 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1141 && a->curr == b->curr
1142 && a->advertised == b->advertised
1143 && a->supported == b->supported
1144 && a->peer == b->peer);
1147 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1148 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1149 * one with the same name or port number). */
1151 ofport_install(struct ofproto *p,
1152 struct netdev *netdev, const struct ofp_phy_port *opp)
1154 const char *netdev_name = netdev_get_name(netdev);
1155 struct ofport *ofport;
1158 /* Create ofport. */
1159 ofport = p->ofproto_class->port_alloc();
1164 ofport->ofproto = p;
1165 ofport->netdev = netdev;
1166 ofport->change_seq = netdev_change_seq(netdev);
1168 ofport->ofp_port = ntohs(opp->port_no);
1170 /* Add port to 'p'. */
1171 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1172 shash_add(&p->port_by_name, netdev_name, ofport);
1174 /* Let the ofproto_class initialize its private data. */
1175 error = p->ofproto_class->port_construct(ofport);
1179 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1183 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1184 p->name, netdev_name, strerror(error));
1186 ofport_destroy__(ofport);
1188 netdev_close(netdev);
1192 /* Removes 'ofport' from 'p' and destroys it. */
1194 ofport_remove(struct ofport *ofport)
1196 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1198 ofport_destroy(ofport);
1201 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1204 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1206 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1208 ofport_remove(port);
1212 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1214 * Does not handle a name or port number change. The caller must implement
1215 * such a change as a delete followed by an add. */
1217 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1219 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1220 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1221 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1222 port->opp.state = opp->state;
1223 port->opp.curr = opp->curr;
1224 port->opp.advertised = opp->advertised;
1225 port->opp.supported = opp->supported;
1226 port->opp.peer = opp->peer;
1228 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1232 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1234 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1236 if (port->ofproto->ofproto_class->set_cfm) {
1237 port->ofproto->ofproto_class->set_cfm(port, NULL);
1239 if (port->ofproto->ofproto_class->bundle_remove) {
1240 port->ofproto->ofproto_class->bundle_remove(port);
1246 ofport_destroy__(struct ofport *port)
1248 struct ofproto *ofproto = port->ofproto;
1249 const char *name = netdev_get_name(port->netdev);
1251 hmap_remove(&ofproto->ports, &port->hmap_node);
1252 shash_delete(&ofproto->port_by_name,
1253 shash_find(&ofproto->port_by_name, name));
1255 netdev_close(port->netdev);
1256 ofproto->ofproto_class->port_dealloc(port);
1260 ofport_destroy(struct ofport *port)
1263 port->ofproto->ofproto_class->port_destruct(port);
1264 ofport_destroy__(port);
1269 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1271 struct ofport *port;
1273 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1274 hash_int(ofp_port, 0), &ofproto->ports) {
1275 if (port->ofp_port == ofp_port) {
1283 update_port(struct ofproto *ofproto, const char *name)
1285 struct ofproto_port ofproto_port;
1286 struct ofp_phy_port opp;
1287 struct netdev *netdev;
1288 struct ofport *port;
1290 COVERAGE_INC(ofproto_update_port);
1292 /* Fetch 'name''s location and properties from the datapath. */
1293 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1294 ? ofport_open(&ofproto_port, &opp)
1297 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1298 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1299 struct netdev *old_netdev = port->netdev;
1301 /* 'name' hasn't changed location. Any properties changed? */
1302 if (!ofport_equal(&port->opp, &opp)) {
1303 ofport_modified(port, &opp);
1306 /* Install the newly opened netdev in case it has changed.
1307 * Don't close the old netdev yet in case port_modified has to
1308 * remove a retained reference to it.*/
1309 port->netdev = netdev;
1310 port->change_seq = netdev_change_seq(netdev);
1312 if (port->ofproto->ofproto_class->port_modified) {
1313 port->ofproto->ofproto_class->port_modified(port);
1316 netdev_close(old_netdev);
1318 /* If 'port' is nonnull then its name differs from 'name' and thus
1319 * we should delete it. If we think there's a port named 'name'
1320 * then its port number must be wrong now so delete it too. */
1322 ofport_remove(port);
1324 ofport_remove_with_name(ofproto, name);
1325 ofport_install(ofproto, netdev, &opp);
1328 /* Any port named 'name' is gone now. */
1329 ofport_remove_with_name(ofproto, name);
1331 ofproto_port_destroy(&ofproto_port);
1335 init_ports(struct ofproto *p)
1337 struct ofproto_port_dump dump;
1338 struct ofproto_port ofproto_port;
1340 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1341 uint16_t ofp_port = ofproto_port.ofp_port;
1342 if (ofproto_get_port(p, ofp_port)) {
1343 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1345 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1346 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1349 struct ofp_phy_port opp;
1350 struct netdev *netdev;
1352 netdev = ofport_open(&ofproto_port, &opp);
1354 ofport_install(p, netdev, &opp);
1363 ofproto_rule_destroy__(struct rule *rule)
1365 free(rule->actions);
1366 rule->ofproto->ofproto_class->rule_dealloc(rule);
1369 /* This function allows an ofproto implementation to destroy any rules that
1370 * remain when its ->destruct() function is called. The caller must have
1371 * already uninitialized any derived members of 'rule' (step 5 described in the
1372 * large comment in ofproto/private.h titled "Life Cycle"). This function
1373 * implements steps 6 and 7.
1375 * This function should only be called from an ofproto implementation's
1376 * ->destruct() function. It is not suitable elsewhere. */
1378 ofproto_rule_destroy(struct rule *rule)
1380 assert(!rule->pending);
1381 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1382 ofproto_rule_destroy__(rule);
1385 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1386 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1389 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1391 const union ofp_action *oa;
1394 if (out_port == OFPP_NONE) {
1397 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1398 if (action_outputs_to_port(oa, htons(out_port))) {
1405 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1406 * statistics appropriately. 'packet' must have at least sizeof(struct
1407 * ofp_packet_in) bytes of headroom.
1409 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1410 * with statistics for 'packet' either way.
1412 * Takes ownership of 'packet'. */
1414 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1418 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1420 flow_extract(packet, 0, in_port, &flow);
1421 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1424 /* Returns true if 'rule' should be hidden from the controller.
1426 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1427 * (e.g. by in-band control) and are intentionally hidden from the
1430 rule_is_hidden(const struct rule *rule)
1432 return rule->cr.priority > UINT16_MAX;
1436 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1438 ofconn_send_reply(ofconn, make_echo_reply(oh));
1443 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1445 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1446 struct ofp_switch_features *osf;
1448 struct ofport *port;
1452 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1453 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1455 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1456 osf->datapath_id = htonll(ofproto->datapath_id);
1457 osf->n_buffers = htonl(pktbuf_capacity());
1458 osf->n_tables = ofproto->n_tables;
1459 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1462 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1464 osf->actions = htonl(actions);
1466 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1467 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1470 ofconn_send_reply(ofconn, buf);
1475 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1477 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1479 struct ofp_switch_config *osc;
1483 /* Figure out flags. */
1484 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1485 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1488 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1489 osc->flags = htons(flags);
1490 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1491 ofconn_send_reply(ofconn, buf);
1497 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1499 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1500 uint16_t flags = ntohs(osc->flags);
1502 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1503 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1504 switch (flags & OFPC_FRAG_MASK) {
1505 case OFPC_FRAG_NORMAL:
1506 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1508 case OFPC_FRAG_DROP:
1509 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1512 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1518 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1523 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1524 * error message code (composed with ofp_mkerr()) for the caller to propagate
1525 * upward. Otherwise, returns 0.
1527 * The log message mentions 'msg_type'. */
1529 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1531 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1532 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1533 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1534 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1537 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1544 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1546 struct ofproto *p = ofconn_get_ofproto(ofconn);
1547 struct ofp_packet_out *opo;
1548 struct ofpbuf payload, *buffer;
1549 union ofp_action *ofp_actions;
1550 struct ofpbuf request;
1552 size_t n_ofp_actions;
1556 COVERAGE_INC(ofproto_packet_out);
1558 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1563 /* Get ofp_packet_out. */
1564 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1565 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1568 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1569 &ofp_actions, &n_ofp_actions);
1575 if (opo->buffer_id != htonl(UINT32_MAX)) {
1576 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1578 if (error || !buffer) {
1587 /* Send out packet. */
1588 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1589 error = p->ofproto_class->packet_out(p, &payload, &flow,
1590 ofp_actions, n_ofp_actions);
1591 ofpbuf_delete(buffer);
1597 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1599 ovs_be32 old_config = port->opp.config;
1601 mask &= config ^ port->opp.config;
1602 if (mask & htonl(OFPPC_PORT_DOWN)) {
1603 if (config & htonl(OFPPC_PORT_DOWN)) {
1604 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1606 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1610 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1611 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1612 OFPPC_NO_PACKET_IN));
1613 if (port->opp.config != old_config) {
1614 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1619 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1621 struct ofproto *p = ofconn_get_ofproto(ofconn);
1622 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1623 struct ofport *port;
1626 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1631 port = ofproto_get_port(p, ntohs(opm->port_no));
1633 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1634 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1635 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1637 update_port_config(port, opm->config, opm->mask);
1638 if (opm->advertise) {
1639 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1646 handle_desc_stats_request(struct ofconn *ofconn,
1647 const struct ofp_stats_msg *request)
1649 struct ofproto *p = ofconn_get_ofproto(ofconn);
1650 struct ofp_desc_stats *ods;
1653 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1654 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1655 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1656 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1657 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1658 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1659 ofconn_send_reply(ofconn, msg);
1665 handle_table_stats_request(struct ofconn *ofconn,
1666 const struct ofp_stats_msg *request)
1668 struct ofproto *p = ofconn_get_ofproto(ofconn);
1669 struct ofp_table_stats *ots;
1673 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1675 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1676 for (i = 0; i < p->n_tables; i++) {
1677 ots[i].table_id = i;
1678 sprintf(ots[i].name, "table%zu", i);
1679 ots[i].wildcards = htonl(OFPFW_ALL);
1680 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1681 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1684 p->ofproto_class->get_tables(p, ots);
1686 ofconn_send_reply(ofconn, msg);
1691 append_port_stat(struct ofport *port, struct list *replies)
1693 struct netdev_stats stats;
1694 struct ofp_port_stats *ops;
1696 /* Intentionally ignore return value, since errors will set
1697 * 'stats' to all-1s, which is correct for OpenFlow, and
1698 * netdev_get_stats() will log errors. */
1699 netdev_get_stats(port->netdev, &stats);
1701 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1702 ops->port_no = port->opp.port_no;
1703 memset(ops->pad, 0, sizeof ops->pad);
1704 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1705 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1706 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1707 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1708 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1709 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1710 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1711 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1712 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1713 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1714 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1715 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1719 handle_port_stats_request(struct ofconn *ofconn,
1720 const struct ofp_port_stats_request *psr)
1722 struct ofproto *p = ofconn_get_ofproto(ofconn);
1723 struct ofport *port;
1724 struct list replies;
1726 ofputil_start_stats_reply(&psr->osm, &replies);
1727 if (psr->port_no != htons(OFPP_NONE)) {
1728 port = ofproto_get_port(p, ntohs(psr->port_no));
1730 append_port_stat(port, &replies);
1733 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1734 append_port_stat(port, &replies);
1738 ofconn_send_replies(ofconn, &replies);
1743 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1745 long long int msecs = time_msec() - start;
1746 *sec = msecs / 1000;
1747 *nsec = (msecs % 1000) * (1000 * 1000);
1750 static struct classifier *
1751 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1753 if (table_id == 0xff) {
1754 return &ofproto->tables[0];
1755 } else if (table_id < ofproto->n_tables) {
1756 return &ofproto->tables[table_id];
1758 /* It would probably be better to reply with an error but there doesn't
1759 * seem to be any appropriate value, so that might just be
1761 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1767 static struct classifier *
1768 next_matching_table(struct ofproto *ofproto,
1769 struct classifier *cls, uint8_t table_id)
1771 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1776 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1779 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1782 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1783 * only once, for that table.
1785 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1786 * and does not enter the loop at all.
1788 * All parameters are evaluated multiple times.
1790 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1791 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1793 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1795 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1796 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1797 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1800 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1801 * to 'out_port' are included.
1803 * Hidden rules are always omitted.
1805 * Returns 0 on success, otherwise an OpenFlow error code. */
1807 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1808 const struct cls_rule *match, uint16_t out_port,
1811 struct classifier *cls;
1814 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1815 struct cls_cursor cursor;
1818 cls_cursor_init(&cursor, cls, match);
1819 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1820 if (rule->pending) {
1821 return OFPROTO_POSTPONE;
1823 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1824 list_push_back(rules, &rule->ofproto_node);
1831 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1832 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1833 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1836 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1837 * to 'out_port' are included.
1839 * Hidden rules are always omitted.
1841 * Returns 0 on success, otherwise an OpenFlow error code. */
1843 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1844 const struct cls_rule *match, uint16_t out_port,
1847 struct classifier *cls;
1850 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1853 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1855 if (rule->pending) {
1856 return OFPROTO_POSTPONE;
1858 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1859 list_push_back(rules, &rule->ofproto_node);
1867 handle_flow_stats_request(struct ofconn *ofconn,
1868 const struct ofp_stats_msg *osm)
1870 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1871 struct flow_stats_request fsr;
1872 struct list replies;
1877 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1882 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1883 fsr.out_port, &rules);
1888 ofputil_start_stats_reply(osm, &replies);
1889 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1890 struct ofputil_flow_stats fs;
1893 fs.cookie = rule->flow_cookie;
1894 fs.table_id = rule->table_id;
1895 calc_flow_duration__(rule->created, &fs.duration_sec,
1897 fs.idle_timeout = rule->idle_timeout;
1898 fs.hard_timeout = rule->hard_timeout;
1899 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1901 fs.actions = rule->actions;
1902 fs.n_actions = rule->n_actions;
1903 ofputil_append_flow_stats_reply(&fs, &replies);
1905 ofconn_send_replies(ofconn, &replies);
1911 flow_stats_ds(struct rule *rule, struct ds *results)
1913 uint64_t packet_count, byte_count;
1915 rule->ofproto->ofproto_class->rule_get_stats(rule,
1916 &packet_count, &byte_count);
1918 if (rule->table_id != 0) {
1919 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1921 ds_put_format(results, "duration=%llds, ",
1922 (time_msec() - rule->created) / 1000);
1923 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1924 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1925 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1926 cls_rule_format(&rule->cr, results);
1927 ds_put_char(results, ',');
1928 if (rule->n_actions > 0) {
1929 ofp_print_actions(results, rule->actions, rule->n_actions);
1931 ds_put_cstr(results, "drop");
1933 ds_put_cstr(results, "\n");
1936 /* Adds a pretty-printed description of all flows to 'results', including
1937 * hidden flows (e.g., set up by in-band control). */
1939 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1941 struct classifier *cls;
1943 for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
1944 struct cls_cursor cursor;
1947 cls_cursor_init(&cursor, cls, NULL);
1948 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1949 flow_stats_ds(rule, results);
1954 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1955 * '*engine_type' and '*engine_id', respectively. */
1957 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1958 uint8_t *engine_type, uint8_t *engine_id)
1960 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1963 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1964 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1965 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
1967 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
1969 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1970 return (ofport && ofproto->ofproto_class->get_cfm_fault
1971 ? ofproto->ofproto_class->get_cfm_fault(ofport)
1976 handle_aggregate_stats_request(struct ofconn *ofconn,
1977 const struct ofp_stats_msg *osm)
1979 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1980 struct flow_stats_request request;
1981 struct ofputil_aggregate_stats stats;
1982 bool unknown_packets, unknown_bytes;
1983 struct ofpbuf *reply;
1988 error = ofputil_decode_flow_stats_request(&request, &osm->header);
1993 error = collect_rules_loose(ofproto, request.table_id, &request.match,
1994 request.out_port, &rules);
1999 memset(&stats, 0, sizeof stats);
2000 unknown_packets = unknown_bytes = false;
2001 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2002 uint64_t packet_count;
2003 uint64_t byte_count;
2005 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2008 if (packet_count == UINT64_MAX) {
2009 unknown_packets = true;
2011 stats.packet_count += packet_count;
2014 if (byte_count == UINT64_MAX) {
2015 unknown_bytes = true;
2017 stats.byte_count += byte_count;
2022 if (unknown_packets) {
2023 stats.packet_count = UINT64_MAX;
2025 if (unknown_bytes) {
2026 stats.byte_count = UINT64_MAX;
2029 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2030 ofconn_send_reply(ofconn, reply);
2035 struct queue_stats_cbdata {
2036 struct ofport *ofport;
2037 struct list replies;
2041 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2042 const struct netdev_queue_stats *stats)
2044 struct ofp_queue_stats *reply;
2046 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2047 reply->port_no = cbdata->ofport->opp.port_no;
2048 memset(reply->pad, 0, sizeof reply->pad);
2049 reply->queue_id = htonl(queue_id);
2050 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2051 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2052 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2056 handle_queue_stats_dump_cb(uint32_t queue_id,
2057 struct netdev_queue_stats *stats,
2060 struct queue_stats_cbdata *cbdata = cbdata_;
2062 put_queue_stats(cbdata, queue_id, stats);
2066 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2067 struct queue_stats_cbdata *cbdata)
2069 cbdata->ofport = port;
2070 if (queue_id == OFPQ_ALL) {
2071 netdev_dump_queue_stats(port->netdev,
2072 handle_queue_stats_dump_cb, cbdata);
2074 struct netdev_queue_stats stats;
2076 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2077 put_queue_stats(cbdata, queue_id, &stats);
2083 handle_queue_stats_request(struct ofconn *ofconn,
2084 const struct ofp_queue_stats_request *qsr)
2086 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2087 struct queue_stats_cbdata cbdata;
2088 struct ofport *port;
2089 unsigned int port_no;
2092 COVERAGE_INC(ofproto_queue_req);
2094 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2096 port_no = ntohs(qsr->port_no);
2097 queue_id = ntohl(qsr->queue_id);
2098 if (port_no == OFPP_ALL) {
2099 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2100 handle_queue_stats_for_port(port, queue_id, &cbdata);
2102 } else if (port_no < OFPP_MAX) {
2103 port = ofproto_get_port(ofproto, port_no);
2105 handle_queue_stats_for_port(port, queue_id, &cbdata);
2108 ofpbuf_list_delete(&cbdata.replies);
2109 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2111 ofconn_send_replies(ofconn, &cbdata.replies);
2117 is_flow_deletion_pending(const struct ofproto *ofproto,
2118 const struct cls_rule *cls_rule,
2121 if (!hmap_is_empty(&ofproto->deletions)) {
2122 struct ofoperation *op;
2124 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2125 cls_rule_hash(cls_rule, table_id),
2126 &ofproto->deletions) {
2127 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2136 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2137 * in which no matching flow already exists in the flow table.
2139 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2140 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2141 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2143 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2146 add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
2147 const struct ofp_header *request)
2149 struct classifier *table;
2150 struct ofopgroup *group;
2151 struct rule *victim;
2155 /* Check for overlap, if requested. */
2156 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2157 struct classifier *cls;
2159 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2160 if (classifier_rule_overlaps(cls, &fm->cr)) {
2161 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2167 if (fm->table_id == 0xff) {
2169 if (ofproto->n_tables > 1) {
2170 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2175 assert(table_id < ofproto->n_tables);
2176 table = &ofproto->tables[table_id];
2178 table = &ofproto->tables[0];
2180 } else if (fm->table_id < ofproto->n_tables) {
2181 table = &ofproto->tables[fm->table_id];
2183 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2186 /* Serialize against pending deletion. */
2187 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2188 return OFPROTO_POSTPONE;
2191 /* Allocate new rule. */
2192 rule = ofproto->ofproto_class->rule_alloc();
2194 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2195 ofproto->name, strerror(error));
2198 rule->ofproto = ofproto;
2200 rule->pending = NULL;
2201 rule->flow_cookie = fm->cookie;
2202 rule->created = time_msec();
2203 rule->idle_timeout = fm->idle_timeout;
2204 rule->hard_timeout = fm->hard_timeout;
2205 rule->table_id = table - ofproto->tables;
2206 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2207 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2208 rule->n_actions = fm->n_actions;
2210 /* Insert new rule. */
2211 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2212 if (victim && victim->pending) {
2213 error = OFPROTO_POSTPONE;
2216 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2217 : ofopgroup_create(ofproto));
2218 ofoperation_create(group, rule, OFOPERATION_ADD);
2219 rule->pending->victim = victim;
2221 error = ofproto->ofproto_class->rule_construct(rule);
2223 ofoperation_destroy(rule->pending);
2225 ofopgroup_submit(group);
2228 /* Back out if an error occurred. */
2231 classifier_replace(table, &victim->cr);
2233 classifier_remove(table, &rule->cr);
2235 ofproto_rule_destroy__(rule);
2240 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2242 /* Modifies the rules listed in 'rules', changing their actions to match those
2245 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2248 * Returns 0 on success, otherwise an OpenFlow error code. */
2250 modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
2251 const struct ofp_header *request, struct list *rules)
2253 struct ofopgroup *group;
2256 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2257 LIST_FOR_EACH (rule, ofproto_node, rules) {
2258 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2259 rule->actions, rule->n_actions)) {
2260 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2261 rule->pending->actions = rule->actions;
2262 rule->pending->n_actions = rule->n_actions;
2263 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2264 rule->n_actions = fm->n_actions;
2265 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2267 rule->flow_cookie = fm->cookie;
2269 ofopgroup_submit(group);
2274 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2275 * encoded by ofp_mkerr() on failure.
2277 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2280 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
2281 const struct ofp_header *request)
2283 struct ofproto *p = ofconn_get_ofproto(ofconn);
2287 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2288 return (error ? error
2289 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2290 : modify_flows__(ofconn, fm, request, &rules));
2293 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2294 * code as encoded by ofp_mkerr() on failure.
2296 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2299 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2300 const struct ofp_header *request)
2302 struct ofproto *p = ofconn_get_ofproto(ofconn);
2306 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2307 return (error ? error
2308 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2309 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2314 /* OFPFC_DELETE implementation. */
2316 /* Deletes the rules listed in 'rules'.
2318 * Returns 0 on success, otherwise an OpenFlow error code. */
2320 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2323 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2324 struct rule *rule, *next;
2325 struct ofopgroup *group;
2327 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2328 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2329 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2331 ofoperation_create(group, rule, OFOPERATION_DELETE);
2332 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2333 rule->ofproto->ofproto_class->rule_destruct(rule);
2335 ofopgroup_submit(group);
2340 /* Implements OFPFC_DELETE. */
2342 delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
2343 const struct ofp_header *request)
2345 struct ofproto *p = ofconn_get_ofproto(ofconn);
2349 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2351 return (error ? error
2352 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2356 /* Implements OFPFC_DELETE_STRICT. */
2358 delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2359 const struct ofp_header *request)
2361 struct ofproto *p = ofconn_get_ofproto(ofconn);
2365 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2367 return (error ? error
2368 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2374 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2376 struct ofputil_flow_removed fr;
2378 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2383 fr.cookie = rule->flow_cookie;
2385 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2386 fr.idle_timeout = rule->idle_timeout;
2387 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2390 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2393 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2394 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2397 * ofproto implementation ->run() functions should use this function to expire
2398 * OpenFlow flows. */
2400 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2402 struct ofproto *ofproto = rule->ofproto;
2403 struct ofopgroup *group;
2405 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2407 ofproto_rule_send_removed(rule, reason);
2409 group = ofopgroup_create(ofproto);
2410 ofoperation_create(group, rule, OFOPERATION_DELETE);
2411 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2412 rule->ofproto->ofproto_class->rule_destruct(rule);
2413 ofopgroup_submit(group);
2417 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2419 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2423 error = reject_slave_controller(ofconn, "flow_mod");
2428 if (list_size(&ofproto->pending) >= 50) {
2429 return OFPROTO_POSTPONE;
2432 error = ofputil_decode_flow_mod(&fm, oh,
2433 ofconn_get_flow_mod_table_id(ofconn));
2438 /* We do not support the emergency flow cache. It will hopefully get
2439 * dropped from OpenFlow in the near future. */
2440 if (fm.flags & OFPFF_EMERG) {
2441 /* There isn't a good fit for an error code, so just state that the
2442 * flow table is full. */
2443 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2446 switch (fm.command) {
2448 return add_flow(ofproto, ofconn, &fm, oh);
2451 return modify_flows_loose(ofconn, &fm, oh);
2453 case OFPFC_MODIFY_STRICT:
2454 return modify_flow_strict(ofconn, &fm, oh);
2457 return delete_flows_loose(ofconn, &fm, oh);
2459 case OFPFC_DELETE_STRICT:
2460 return delete_flow_strict(ofconn, &fm, oh);
2463 if (fm.command > 0xff) {
2464 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2465 "flow_mod_table_id extension is not enabled");
2467 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2472 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2474 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2475 struct nx_role_request *reply;
2479 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2480 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2481 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2484 role = ntohl(nrr->role);
2485 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2486 && role != NX_ROLE_SLAVE) {
2487 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2489 /* There's no good error code for this. */
2490 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2493 if (ofconn_get_role(ofconn) != role
2494 && ofconn_has_pending_opgroups(ofconn)) {
2495 return OFPROTO_POSTPONE;
2498 ofconn_set_role(ofconn, role);
2500 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2501 reply->role = htonl(role);
2502 ofconn_send_reply(ofconn, buf);
2508 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2509 const struct ofp_header *oh)
2511 const struct nxt_flow_mod_table_id *msg
2512 = (const struct nxt_flow_mod_table_id *) oh;
2514 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2519 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2521 const struct nxt_set_flow_format *msg
2522 = (const struct nxt_set_flow_format *) oh;
2525 format = ntohl(msg->format);
2526 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2527 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2530 if (format != ofconn_get_flow_format(ofconn)
2531 && ofconn_has_pending_opgroups(ofconn)) {
2532 /* Avoid sending async messages in surprising flow format. */
2533 return OFPROTO_POSTPONE;
2536 ofconn_set_flow_format(ofconn, format);
2541 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2543 struct ofp_header *ob;
2546 if (ofconn_has_pending_opgroups(ofconn)) {
2547 return OFPROTO_POSTPONE;
2550 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2551 ofconn_send_reply(ofconn, buf);
2556 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2558 const struct ofp_header *oh = msg->data;
2559 const struct ofputil_msg_type *type;
2562 error = ofputil_decode_msg_type(oh, &type);
2567 switch (ofputil_msg_type_code(type)) {
2568 /* OpenFlow requests. */
2569 case OFPUTIL_OFPT_ECHO_REQUEST:
2570 return handle_echo_request(ofconn, oh);
2572 case OFPUTIL_OFPT_FEATURES_REQUEST:
2573 return handle_features_request(ofconn, oh);
2575 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2576 return handle_get_config_request(ofconn, oh);
2578 case OFPUTIL_OFPT_SET_CONFIG:
2579 return handle_set_config(ofconn, msg->data);
2581 case OFPUTIL_OFPT_PACKET_OUT:
2582 return handle_packet_out(ofconn, oh);
2584 case OFPUTIL_OFPT_PORT_MOD:
2585 return handle_port_mod(ofconn, oh);
2587 case OFPUTIL_OFPT_FLOW_MOD:
2588 return handle_flow_mod(ofconn, oh);
2590 case OFPUTIL_OFPT_BARRIER_REQUEST:
2591 return handle_barrier_request(ofconn, oh);
2593 /* OpenFlow replies. */
2594 case OFPUTIL_OFPT_ECHO_REPLY:
2597 /* Nicira extension requests. */
2598 case OFPUTIL_NXT_ROLE_REQUEST:
2599 return handle_role_request(ofconn, oh);
2601 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2602 return handle_nxt_flow_mod_table_id(ofconn, oh);
2604 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2605 return handle_nxt_set_flow_format(ofconn, oh);
2607 case OFPUTIL_NXT_FLOW_MOD:
2608 return handle_flow_mod(ofconn, oh);
2610 /* Statistics requests. */
2611 case OFPUTIL_OFPST_DESC_REQUEST:
2612 return handle_desc_stats_request(ofconn, msg->data);
2614 case OFPUTIL_OFPST_FLOW_REQUEST:
2615 case OFPUTIL_NXST_FLOW_REQUEST:
2616 return handle_flow_stats_request(ofconn, msg->data);
2618 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2619 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2620 return handle_aggregate_stats_request(ofconn, msg->data);
2622 case OFPUTIL_OFPST_TABLE_REQUEST:
2623 return handle_table_stats_request(ofconn, msg->data);
2625 case OFPUTIL_OFPST_PORT_REQUEST:
2626 return handle_port_stats_request(ofconn, msg->data);
2628 case OFPUTIL_OFPST_QUEUE_REQUEST:
2629 return handle_queue_stats_request(ofconn, msg->data);
2631 case OFPUTIL_MSG_INVALID:
2632 case OFPUTIL_OFPT_HELLO:
2633 case OFPUTIL_OFPT_ERROR:
2634 case OFPUTIL_OFPT_FEATURES_REPLY:
2635 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2636 case OFPUTIL_OFPT_PACKET_IN:
2637 case OFPUTIL_OFPT_FLOW_REMOVED:
2638 case OFPUTIL_OFPT_PORT_STATUS:
2639 case OFPUTIL_OFPT_BARRIER_REPLY:
2640 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2641 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2642 case OFPUTIL_OFPST_DESC_REPLY:
2643 case OFPUTIL_OFPST_FLOW_REPLY:
2644 case OFPUTIL_OFPST_QUEUE_REPLY:
2645 case OFPUTIL_OFPST_PORT_REPLY:
2646 case OFPUTIL_OFPST_TABLE_REPLY:
2647 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2648 case OFPUTIL_NXT_ROLE_REPLY:
2649 case OFPUTIL_NXT_FLOW_REMOVED:
2650 case OFPUTIL_NXST_FLOW_REPLY:
2651 case OFPUTIL_NXST_AGGREGATE_REPLY:
2653 if (VLOG_IS_WARN_ENABLED()) {
2654 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2655 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2658 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2659 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2661 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2667 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2669 int error = handle_openflow__(ofconn, ofp_msg);
2670 if (error && error != OFPROTO_POSTPONE) {
2671 ofconn_send_error(ofconn, ofp_msg->data, error);
2673 COVERAGE_INC(ofproto_recv_openflow);
2674 return error != OFPROTO_POSTPONE;
2677 /* Asynchronous operations. */
2679 /* Creates and returns a new ofopgroup that is not associated with any
2680 * OpenFlow connection.
2682 * The caller should add operations to the returned group with
2683 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2684 static struct ofopgroup *
2685 ofopgroup_create(struct ofproto *ofproto)
2687 struct ofopgroup *group = xzalloc(sizeof *group);
2688 group->ofproto = ofproto;
2689 list_init(&group->ofproto_node);
2690 list_init(&group->ops);
2691 list_init(&group->ofconn_node);
2695 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2696 * the ofopgroup eventually fails, then the error reply will include 'request'.
2697 * If the ofopgroup eventually succeeds, then the packet with buffer id
2698 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2700 * The caller should add operations to the returned group with
2701 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2702 static struct ofopgroup *
2703 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2704 const struct ofp_header *request,
2707 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2708 size_t request_len = ntohs(request->length);
2710 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2711 group->ofconn = ofconn;
2712 group->request = xmemdup(request, MIN(request_len, 64));
2713 group->buffer_id = buffer_id;
2718 /* Submits 'group' for processing.
2720 * If 'group' contains no operations (e.g. none were ever added, or all of the
2721 * ones that were added completed synchronously), then it is destroyed
2722 * immediately. Otherwise it is added to the ofproto's list of pending
2725 ofopgroup_submit(struct ofopgroup *group)
2727 if (list_is_empty(&group->ops)) {
2728 ofopgroup_destroy(group);
2730 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2735 ofopgroup_destroy(struct ofopgroup *group)
2737 assert(list_is_empty(&group->ops));
2738 if (!list_is_empty(&group->ofproto_node)) {
2739 list_remove(&group->ofproto_node);
2741 if (!list_is_empty(&group->ofconn_node)) {
2742 list_remove(&group->ofconn_node);
2744 ofconn_send_error(group->ofconn, group->request, group->error);
2746 connmgr_retry(group->ofproto->connmgr);
2748 free(group->request);
2752 /* Initiates a new operation on 'rule', of the specified 'type', within
2753 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2755 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2756 enum ofoperation_type type)
2758 struct ofoperation *op;
2760 assert(!rule->pending);
2762 op = rule->pending = xzalloc(sizeof *op);
2764 list_push_back(&group->ops, &op->group_node);
2768 op->flow_cookie = rule->flow_cookie;
2770 if (type == OFOPERATION_DELETE) {
2771 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2772 cls_rule_hash(&rule->cr, rule->table_id));
2777 ofoperation_destroy(struct ofoperation *op)
2779 struct ofopgroup *group = op->group;
2782 op->rule->pending = NULL;
2784 if (op->type == OFOPERATION_DELETE) {
2785 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2787 list_remove(&op->group_node);
2791 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2792 ofopgroup_destroy(group);
2796 /* Indicates that 'op' completed with status 'error', which is either 0 to
2797 * indicate success or an OpenFlow error code (constructed with
2798 * e.g. ofp_mkerr()).
2800 * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
2801 * deletions are not allowed to fail.
2803 * Please see the large comment in ofproto/private.h titled "Asynchronous
2804 * Operation Support" for more information. */
2806 ofoperation_complete(struct ofoperation *op, int error)
2808 struct ofopgroup *group = op->group;
2809 struct rule *rule = op->rule;
2810 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2812 assert(rule->pending == op);
2813 assert(op->status < 0);
2818 && op->type != OFOPERATION_DELETE
2820 && group->buffer_id != UINT32_MAX
2821 && list_is_singleton(&op->group_node)) {
2822 struct ofpbuf *packet;
2825 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2829 error = rule_execute(rule, in_port, packet);
2832 if (!group->error) {
2833 group->error = error;
2837 case OFOPERATION_ADD:
2840 ofproto_rule_destroy__(op->victim);
2844 classifier_replace(table, &op->victim->cr);
2847 classifier_remove(table, &rule->cr);
2849 ofproto_rule_destroy__(rule);
2854 case OFOPERATION_DELETE:
2856 ofproto_rule_destroy__(rule);
2860 case OFOPERATION_MODIFY:
2862 free(rule->actions);
2863 rule->actions = op->actions;
2864 rule->n_actions = op->n_actions;
2872 ofoperation_destroy(op);
2876 ofoperation_get_victim(struct ofoperation *op)
2878 assert(op->type == OFOPERATION_ADD);
2883 pick_datapath_id(const struct ofproto *ofproto)
2885 const struct ofport *port;
2887 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2889 uint8_t ea[ETH_ADDR_LEN];
2892 error = netdev_get_etheraddr(port->netdev, ea);
2894 return eth_addr_to_uint64(ea);
2896 VLOG_WARN("could not get MAC address for %s (%s)",
2897 netdev_get_name(port->netdev), strerror(error));
2899 return ofproto->fallback_dpid;
2903 pick_fallback_dpid(void)
2905 uint8_t ea[ETH_ADDR_LEN];
2906 eth_addr_nicira_random(ea);
2907 return eth_addr_to_uint64(ea);
2910 /* unixctl commands. */
2913 ofproto_lookup(const char *name)
2915 struct ofproto *ofproto;
2917 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2919 if (!strcmp(ofproto->name, name)) {
2927 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2928 void *aux OVS_UNUSED)
2930 struct ofproto *ofproto;
2934 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2935 ds_put_format(&results, "%s\n", ofproto->name);
2937 unixctl_command_reply(conn, 200, ds_cstr(&results));
2938 ds_destroy(&results);
2942 ofproto_unixctl_init(void)
2944 static bool registered;
2950 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);