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 struct shash_node *node;
850 SHASH_FOR_EACH (node, info) {
851 struct ofproto_controller_info *cinfo = node->data;
852 while (cinfo->pairs.n) {
853 free((char *) cinfo->pairs.values[--cinfo->pairs.n]);
860 /* Makes a deep copy of 'old' into 'port'. */
862 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
864 port->name = xstrdup(old->name);
865 port->type = xstrdup(old->type);
866 port->ofp_port = old->ofp_port;
869 /* Frees memory allocated to members of 'ofproto_port'.
871 * Do not call this function on an ofproto_port obtained from
872 * ofproto_port_dump_next(): that function retains ownership of the data in the
875 ofproto_port_destroy(struct ofproto_port *ofproto_port)
877 free(ofproto_port->name);
878 free(ofproto_port->type);
881 /* Initializes 'dump' to begin dumping the ports in an ofproto.
883 * This function provides no status indication. An error status for the entire
884 * dump operation is provided when it is completed by calling
885 * ofproto_port_dump_done().
888 ofproto_port_dump_start(struct ofproto_port_dump *dump,
889 const struct ofproto *ofproto)
891 dump->ofproto = ofproto;
892 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
896 /* Attempts to retrieve another port from 'dump', which must have been created
897 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
898 * 'port' and returns true. On failure, returns false.
900 * Failure might indicate an actual error or merely that the last port has been
901 * dumped. An error status for the entire dump operation is provided when it
902 * is completed by calling ofproto_port_dump_done().
904 * The ofproto owns the data stored in 'port'. It will remain valid until at
905 * least the next time 'dump' is passed to ofproto_port_dump_next() or
906 * ofproto_port_dump_done(). */
908 ofproto_port_dump_next(struct ofproto_port_dump *dump,
909 struct ofproto_port *port)
911 const struct ofproto *ofproto = dump->ofproto;
917 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
920 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
926 /* Completes port table dump operation 'dump', which must have been created
927 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
928 * error-free, otherwise a positive errno value describing the problem. */
930 ofproto_port_dump_done(struct ofproto_port_dump *dump)
932 const struct ofproto *ofproto = dump->ofproto;
934 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
937 return dump->error == EOF ? 0 : dump->error;
940 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
941 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
942 * is non-null). On failure, returns a positive errno value and sets
943 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
945 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
951 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
953 update_port(ofproto, netdev_get_name(netdev));
956 *ofp_portp = error ? OFPP_NONE : ofp_port;
961 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
962 * initializes '*port' appropriately; on failure, returns a positive errno
965 * The caller owns the data in 'ofproto_port' and must free it with
966 * ofproto_port_destroy() when it is no longer needed. */
968 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
969 struct ofproto_port *port)
973 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
975 memset(port, 0, sizeof *port);
980 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
981 * Returns 0 if successful, otherwise a positive errno. */
983 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
985 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
986 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
989 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
990 if (!error && ofport) {
991 /* 'name' is the netdev's name and update_port() is going to close the
992 * netdev. Just in case update_port() refers to 'name' after it
993 * destroys 'ofport', make a copy of it around the update_port()
995 char *devname = xstrdup(name);
996 update_port(ofproto, devname);
1002 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1003 * performs the 'n_actions' actions in 'actions'. The new flow will not
1006 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1007 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1008 * controllers; otherwise, it will be hidden.
1010 * The caller retains ownership of 'cls_rule' and 'actions'.
1012 * This is a helper function for in-band control and fail-open. */
1014 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1015 const union ofp_action *actions, size_t n_actions)
1017 const struct rule *rule;
1019 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1020 &ofproto->tables[0], cls_rule));
1021 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1022 actions, n_actions)) {
1025 memset(&fm, 0, sizeof fm);
1027 fm.buffer_id = UINT32_MAX;
1028 fm.actions = (union ofp_action *) actions;
1029 fm.n_actions = n_actions;
1030 add_flow(ofproto, NULL, &fm, NULL);
1034 /* Searches for a rule with matching criteria exactly equal to 'target' in
1035 * ofproto's table 0 and, if it finds one, deletes it.
1037 * This is a helper function for in-band control and fail-open. */
1039 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1043 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1044 &ofproto->tables[0], target));
1046 /* No such rule -> success. */
1048 } else if (rule->pending) {
1049 /* An operation on the rule is already pending -> failure.
1050 * Caller must retry later if it's important. */
1053 /* Initiate deletion -> success. */
1054 struct ofopgroup *group = ofopgroup_create(ofproto);
1055 ofoperation_create(group, rule, OFOPERATION_DELETE);
1056 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1057 rule->ofproto->ofproto_class->rule_destruct(rule);
1058 ofopgroup_submit(group);
1064 /* Starts the process of deleting all of the flows from all of ofproto's flow
1065 * tables and then reintroducing the flows required by in-band control and
1066 * fail-open. The process will complete in a later call to ofproto_run(). */
1068 ofproto_flush_flows(struct ofproto *ofproto)
1070 COVERAGE_INC(ofproto_flush);
1071 ofproto->state = S_FLUSH;
1075 reinit_ports(struct ofproto *p)
1077 struct ofproto_port_dump dump;
1078 struct sset devnames;
1079 struct ofport *ofport;
1080 struct ofproto_port ofproto_port;
1081 const char *devname;
1083 COVERAGE_INC(ofproto_reinit_ports);
1085 sset_init(&devnames);
1086 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1087 sset_add(&devnames, netdev_get_name(ofport->netdev));
1089 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1090 sset_add(&devnames, ofproto_port.name);
1093 SSET_FOR_EACH (devname, &devnames) {
1094 update_port(p, devname);
1096 sset_destroy(&devnames);
1099 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1100 * netdev cannot be opened. On success, also fills in 'opp'. */
1101 static struct netdev *
1102 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1104 uint32_t curr, advertised, supported, peer;
1105 struct netdev_options netdev_options;
1106 enum netdev_flags flags;
1107 struct netdev *netdev;
1110 memset(&netdev_options, 0, sizeof netdev_options);
1111 netdev_options.name = ofproto_port->name;
1112 netdev_options.type = ofproto_port->type;
1113 netdev_options.ethertype = NETDEV_ETH_TYPE_NONE;
1115 error = netdev_open(&netdev_options, &netdev);
1117 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1118 "cannot be opened (%s)",
1119 ofproto_port->name, ofproto_port->ofp_port,
1120 ofproto_port->name, strerror(error));
1124 netdev_get_flags(netdev, &flags);
1125 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1127 opp->port_no = htons(ofproto_port->ofp_port);
1128 netdev_get_etheraddr(netdev, opp->hw_addr);
1129 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1130 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1131 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1132 opp->curr = htonl(curr);
1133 opp->advertised = htonl(advertised);
1134 opp->supported = htonl(supported);
1135 opp->peer = htonl(peer);
1140 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1141 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1144 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1146 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1147 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1148 && a->state == b->state
1149 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1150 && a->curr == b->curr
1151 && a->advertised == b->advertised
1152 && a->supported == b->supported
1153 && a->peer == b->peer);
1156 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1157 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1158 * one with the same name or port number). */
1160 ofport_install(struct ofproto *p,
1161 struct netdev *netdev, const struct ofp_phy_port *opp)
1163 const char *netdev_name = netdev_get_name(netdev);
1164 struct ofport *ofport;
1167 /* Create ofport. */
1168 ofport = p->ofproto_class->port_alloc();
1173 ofport->ofproto = p;
1174 ofport->netdev = netdev;
1175 ofport->change_seq = netdev_change_seq(netdev);
1177 ofport->ofp_port = ntohs(opp->port_no);
1179 /* Add port to 'p'. */
1180 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1181 shash_add(&p->port_by_name, netdev_name, ofport);
1183 /* Let the ofproto_class initialize its private data. */
1184 error = p->ofproto_class->port_construct(ofport);
1188 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1192 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1193 p->name, netdev_name, strerror(error));
1195 ofport_destroy__(ofport);
1197 netdev_close(netdev);
1201 /* Removes 'ofport' from 'p' and destroys it. */
1203 ofport_remove(struct ofport *ofport)
1205 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1207 ofport_destroy(ofport);
1210 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1213 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1215 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1217 ofport_remove(port);
1221 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1223 * Does not handle a name or port number change. The caller must implement
1224 * such a change as a delete followed by an add. */
1226 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1228 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1229 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1230 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1231 port->opp.state = opp->state;
1232 port->opp.curr = opp->curr;
1233 port->opp.advertised = opp->advertised;
1234 port->opp.supported = opp->supported;
1235 port->opp.peer = opp->peer;
1237 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1241 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1243 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1245 if (port->ofproto->ofproto_class->set_cfm) {
1246 port->ofproto->ofproto_class->set_cfm(port, NULL);
1248 if (port->ofproto->ofproto_class->bundle_remove) {
1249 port->ofproto->ofproto_class->bundle_remove(port);
1255 ofport_destroy__(struct ofport *port)
1257 struct ofproto *ofproto = port->ofproto;
1258 const char *name = netdev_get_name(port->netdev);
1260 hmap_remove(&ofproto->ports, &port->hmap_node);
1261 shash_delete(&ofproto->port_by_name,
1262 shash_find(&ofproto->port_by_name, name));
1264 netdev_close(port->netdev);
1265 ofproto->ofproto_class->port_dealloc(port);
1269 ofport_destroy(struct ofport *port)
1272 port->ofproto->ofproto_class->port_destruct(port);
1273 ofport_destroy__(port);
1278 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1280 struct ofport *port;
1282 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1283 hash_int(ofp_port, 0), &ofproto->ports) {
1284 if (port->ofp_port == ofp_port) {
1292 update_port(struct ofproto *ofproto, const char *name)
1294 struct ofproto_port ofproto_port;
1295 struct ofp_phy_port opp;
1296 struct netdev *netdev;
1297 struct ofport *port;
1299 COVERAGE_INC(ofproto_update_port);
1301 /* Fetch 'name''s location and properties from the datapath. */
1302 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1303 ? ofport_open(&ofproto_port, &opp)
1306 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1307 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1308 struct netdev *old_netdev = port->netdev;
1310 /* 'name' hasn't changed location. Any properties changed? */
1311 if (!ofport_equal(&port->opp, &opp)) {
1312 ofport_modified(port, &opp);
1315 /* Install the newly opened netdev in case it has changed.
1316 * Don't close the old netdev yet in case port_modified has to
1317 * remove a retained reference to it.*/
1318 port->netdev = netdev;
1319 port->change_seq = netdev_change_seq(netdev);
1321 if (port->ofproto->ofproto_class->port_modified) {
1322 port->ofproto->ofproto_class->port_modified(port);
1325 netdev_close(old_netdev);
1327 /* If 'port' is nonnull then its name differs from 'name' and thus
1328 * we should delete it. If we think there's a port named 'name'
1329 * then its port number must be wrong now so delete it too. */
1331 ofport_remove(port);
1333 ofport_remove_with_name(ofproto, name);
1334 ofport_install(ofproto, netdev, &opp);
1337 /* Any port named 'name' is gone now. */
1338 ofport_remove_with_name(ofproto, name);
1340 ofproto_port_destroy(&ofproto_port);
1344 init_ports(struct ofproto *p)
1346 struct ofproto_port_dump dump;
1347 struct ofproto_port ofproto_port;
1349 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1350 uint16_t ofp_port = ofproto_port.ofp_port;
1351 if (ofproto_get_port(p, ofp_port)) {
1352 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1354 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1355 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1358 struct ofp_phy_port opp;
1359 struct netdev *netdev;
1361 netdev = ofport_open(&ofproto_port, &opp);
1363 ofport_install(p, netdev, &opp);
1372 ofproto_rule_destroy__(struct rule *rule)
1374 free(rule->actions);
1375 rule->ofproto->ofproto_class->rule_dealloc(rule);
1378 /* This function allows an ofproto implementation to destroy any rules that
1379 * remain when its ->destruct() function is called. The caller must have
1380 * already uninitialized any derived members of 'rule' (step 5 described in the
1381 * large comment in ofproto/private.h titled "Life Cycle"). This function
1382 * implements steps 6 and 7.
1384 * This function should only be called from an ofproto implementation's
1385 * ->destruct() function. It is not suitable elsewhere. */
1387 ofproto_rule_destroy(struct rule *rule)
1389 assert(!rule->pending);
1390 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1391 ofproto_rule_destroy__(rule);
1394 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1395 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1398 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1400 const union ofp_action *oa;
1403 if (out_port == OFPP_NONE) {
1406 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1407 if (action_outputs_to_port(oa, htons(out_port))) {
1414 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1415 * statistics appropriately. 'packet' must have at least sizeof(struct
1416 * ofp_packet_in) bytes of headroom.
1418 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1419 * with statistics for 'packet' either way.
1421 * Takes ownership of 'packet'. */
1423 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1427 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1429 flow_extract(packet, 0, in_port, &flow);
1430 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1433 /* Returns true if 'rule' should be hidden from the controller.
1435 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1436 * (e.g. by in-band control) and are intentionally hidden from the
1439 rule_is_hidden(const struct rule *rule)
1441 return rule->cr.priority > UINT16_MAX;
1445 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1447 ofconn_send_reply(ofconn, make_echo_reply(oh));
1452 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1454 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1455 struct ofp_switch_features *osf;
1457 struct ofport *port;
1461 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1462 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1464 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1465 osf->datapath_id = htonll(ofproto->datapath_id);
1466 osf->n_buffers = htonl(pktbuf_capacity());
1467 osf->n_tables = ofproto->n_tables;
1468 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1471 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1473 osf->actions = htonl(actions);
1475 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1476 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1479 ofconn_send_reply(ofconn, buf);
1484 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1486 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1488 struct ofp_switch_config *osc;
1492 /* Figure out flags. */
1493 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1494 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1497 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1498 osc->flags = htons(flags);
1499 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1500 ofconn_send_reply(ofconn, buf);
1506 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1508 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1509 uint16_t flags = ntohs(osc->flags);
1511 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1512 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1513 switch (flags & OFPC_FRAG_MASK) {
1514 case OFPC_FRAG_NORMAL:
1515 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1517 case OFPC_FRAG_DROP:
1518 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1521 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1527 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1532 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1533 * error message code (composed with ofp_mkerr()) for the caller to propagate
1534 * upward. Otherwise, returns 0.
1536 * The log message mentions 'msg_type'. */
1538 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1540 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1541 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1542 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1543 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1546 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1553 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1555 struct ofproto *p = ofconn_get_ofproto(ofconn);
1556 struct ofp_packet_out *opo;
1557 struct ofpbuf payload, *buffer;
1558 union ofp_action *ofp_actions;
1559 struct ofpbuf request;
1561 size_t n_ofp_actions;
1565 COVERAGE_INC(ofproto_packet_out);
1567 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1572 /* Get ofp_packet_out. */
1573 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1574 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1577 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1578 &ofp_actions, &n_ofp_actions);
1584 if (opo->buffer_id != htonl(UINT32_MAX)) {
1585 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1587 if (error || !buffer) {
1596 /* Send out packet. */
1597 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1598 error = p->ofproto_class->packet_out(p, &payload, &flow,
1599 ofp_actions, n_ofp_actions);
1600 ofpbuf_delete(buffer);
1606 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1608 ovs_be32 old_config = port->opp.config;
1610 mask &= config ^ port->opp.config;
1611 if (mask & htonl(OFPPC_PORT_DOWN)) {
1612 if (config & htonl(OFPPC_PORT_DOWN)) {
1613 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1615 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1619 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1620 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1621 OFPPC_NO_PACKET_IN));
1622 if (port->opp.config != old_config) {
1623 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1628 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1630 struct ofproto *p = ofconn_get_ofproto(ofconn);
1631 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1632 struct ofport *port;
1635 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1640 port = ofproto_get_port(p, ntohs(opm->port_no));
1642 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1643 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1644 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1646 update_port_config(port, opm->config, opm->mask);
1647 if (opm->advertise) {
1648 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1655 handle_desc_stats_request(struct ofconn *ofconn,
1656 const struct ofp_stats_msg *request)
1658 struct ofproto *p = ofconn_get_ofproto(ofconn);
1659 struct ofp_desc_stats *ods;
1662 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1663 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1664 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1665 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1666 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1667 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1668 ofconn_send_reply(ofconn, msg);
1674 handle_table_stats_request(struct ofconn *ofconn,
1675 const struct ofp_stats_msg *request)
1677 struct ofproto *p = ofconn_get_ofproto(ofconn);
1678 struct ofp_table_stats *ots;
1682 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1684 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1685 for (i = 0; i < p->n_tables; i++) {
1686 ots[i].table_id = i;
1687 sprintf(ots[i].name, "table%zu", i);
1688 ots[i].wildcards = htonl(OFPFW_ALL);
1689 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1690 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1693 p->ofproto_class->get_tables(p, ots);
1695 ofconn_send_reply(ofconn, msg);
1700 append_port_stat(struct ofport *port, struct list *replies)
1702 struct netdev_stats stats;
1703 struct ofp_port_stats *ops;
1705 /* Intentionally ignore return value, since errors will set
1706 * 'stats' to all-1s, which is correct for OpenFlow, and
1707 * netdev_get_stats() will log errors. */
1708 netdev_get_stats(port->netdev, &stats);
1710 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1711 ops->port_no = port->opp.port_no;
1712 memset(ops->pad, 0, sizeof ops->pad);
1713 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1714 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1715 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1716 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1717 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1718 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1719 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1720 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1721 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1722 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1723 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1724 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1728 handle_port_stats_request(struct ofconn *ofconn,
1729 const struct ofp_port_stats_request *psr)
1731 struct ofproto *p = ofconn_get_ofproto(ofconn);
1732 struct ofport *port;
1733 struct list replies;
1735 ofputil_start_stats_reply(&psr->osm, &replies);
1736 if (psr->port_no != htons(OFPP_NONE)) {
1737 port = ofproto_get_port(p, ntohs(psr->port_no));
1739 append_port_stat(port, &replies);
1742 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1743 append_port_stat(port, &replies);
1747 ofconn_send_replies(ofconn, &replies);
1752 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1754 long long int msecs = time_msec() - start;
1755 *sec = msecs / 1000;
1756 *nsec = (msecs % 1000) * (1000 * 1000);
1759 static struct classifier *
1760 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1762 if (table_id == 0xff) {
1763 return &ofproto->tables[0];
1764 } else if (table_id < ofproto->n_tables) {
1765 return &ofproto->tables[table_id];
1767 /* It would probably be better to reply with an error but there doesn't
1768 * seem to be any appropriate value, so that might just be
1770 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1776 static struct classifier *
1777 next_matching_table(struct ofproto *ofproto,
1778 struct classifier *cls, uint8_t table_id)
1780 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1785 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1788 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1791 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1792 * only once, for that table.
1794 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1795 * and does not enter the loop at all.
1797 * All parameters are evaluated multiple times.
1799 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1800 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1802 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1804 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1805 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1806 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1809 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1810 * to 'out_port' are included.
1812 * Hidden rules are always omitted.
1814 * Returns 0 on success, otherwise an OpenFlow error code. */
1816 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1817 const struct cls_rule *match, uint16_t out_port,
1820 struct classifier *cls;
1823 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1824 struct cls_cursor cursor;
1827 cls_cursor_init(&cursor, cls, match);
1828 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1829 if (rule->pending) {
1830 return OFPROTO_POSTPONE;
1832 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1833 list_push_back(rules, &rule->ofproto_node);
1840 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1841 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1842 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1845 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1846 * to 'out_port' are included.
1848 * Hidden rules are always omitted.
1850 * Returns 0 on success, otherwise an OpenFlow error code. */
1852 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1853 const struct cls_rule *match, uint16_t out_port,
1856 struct classifier *cls;
1859 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1862 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1864 if (rule->pending) {
1865 return OFPROTO_POSTPONE;
1867 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1868 list_push_back(rules, &rule->ofproto_node);
1876 handle_flow_stats_request(struct ofconn *ofconn,
1877 const struct ofp_stats_msg *osm)
1879 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1880 struct flow_stats_request fsr;
1881 struct list replies;
1886 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1891 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1892 fsr.out_port, &rules);
1897 ofputil_start_stats_reply(osm, &replies);
1898 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1899 struct ofputil_flow_stats fs;
1902 fs.cookie = rule->flow_cookie;
1903 fs.table_id = rule->table_id;
1904 calc_flow_duration__(rule->created, &fs.duration_sec,
1906 fs.idle_timeout = rule->idle_timeout;
1907 fs.hard_timeout = rule->hard_timeout;
1908 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1910 fs.actions = rule->actions;
1911 fs.n_actions = rule->n_actions;
1912 ofputil_append_flow_stats_reply(&fs, &replies);
1914 ofconn_send_replies(ofconn, &replies);
1920 flow_stats_ds(struct rule *rule, struct ds *results)
1922 uint64_t packet_count, byte_count;
1924 rule->ofproto->ofproto_class->rule_get_stats(rule,
1925 &packet_count, &byte_count);
1927 if (rule->table_id != 0) {
1928 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1930 ds_put_format(results, "duration=%llds, ",
1931 (time_msec() - rule->created) / 1000);
1932 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1933 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1934 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1935 cls_rule_format(&rule->cr, results);
1936 ds_put_char(results, ',');
1937 if (rule->n_actions > 0) {
1938 ofp_print_actions(results, rule->actions, rule->n_actions);
1940 ds_put_cstr(results, "drop");
1942 ds_put_cstr(results, "\n");
1945 /* Adds a pretty-printed description of all flows to 'results', including
1946 * hidden flows (e.g., set up by in-band control). */
1948 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1950 struct classifier *cls;
1952 for (cls = &p->tables[0]; cls < &p->tables[p->n_tables]; cls++) {
1953 struct cls_cursor cursor;
1956 cls_cursor_init(&cursor, cls, NULL);
1957 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1958 flow_stats_ds(rule, results);
1963 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1964 * '*engine_type' and '*engine_id', respectively. */
1966 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1967 uint8_t *engine_type, uint8_t *engine_id)
1969 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1972 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1973 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1974 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
1976 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
1978 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1979 return (ofport && ofproto->ofproto_class->get_cfm_fault
1980 ? ofproto->ofproto_class->get_cfm_fault(ofport)
1985 handle_aggregate_stats_request(struct ofconn *ofconn,
1986 const struct ofp_stats_msg *osm)
1988 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1989 struct flow_stats_request request;
1990 struct ofputil_aggregate_stats stats;
1991 bool unknown_packets, unknown_bytes;
1992 struct ofpbuf *reply;
1997 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2002 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2003 request.out_port, &rules);
2008 memset(&stats, 0, sizeof stats);
2009 unknown_packets = unknown_bytes = false;
2010 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2011 uint64_t packet_count;
2012 uint64_t byte_count;
2014 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2017 if (packet_count == UINT64_MAX) {
2018 unknown_packets = true;
2020 stats.packet_count += packet_count;
2023 if (byte_count == UINT64_MAX) {
2024 unknown_bytes = true;
2026 stats.byte_count += byte_count;
2031 if (unknown_packets) {
2032 stats.packet_count = UINT64_MAX;
2034 if (unknown_bytes) {
2035 stats.byte_count = UINT64_MAX;
2038 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2039 ofconn_send_reply(ofconn, reply);
2044 struct queue_stats_cbdata {
2045 struct ofport *ofport;
2046 struct list replies;
2050 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2051 const struct netdev_queue_stats *stats)
2053 struct ofp_queue_stats *reply;
2055 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2056 reply->port_no = cbdata->ofport->opp.port_no;
2057 memset(reply->pad, 0, sizeof reply->pad);
2058 reply->queue_id = htonl(queue_id);
2059 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2060 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2061 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2065 handle_queue_stats_dump_cb(uint32_t queue_id,
2066 struct netdev_queue_stats *stats,
2069 struct queue_stats_cbdata *cbdata = cbdata_;
2071 put_queue_stats(cbdata, queue_id, stats);
2075 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2076 struct queue_stats_cbdata *cbdata)
2078 cbdata->ofport = port;
2079 if (queue_id == OFPQ_ALL) {
2080 netdev_dump_queue_stats(port->netdev,
2081 handle_queue_stats_dump_cb, cbdata);
2083 struct netdev_queue_stats stats;
2085 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2086 put_queue_stats(cbdata, queue_id, &stats);
2092 handle_queue_stats_request(struct ofconn *ofconn,
2093 const struct ofp_queue_stats_request *qsr)
2095 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2096 struct queue_stats_cbdata cbdata;
2097 struct ofport *port;
2098 unsigned int port_no;
2101 COVERAGE_INC(ofproto_queue_req);
2103 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2105 port_no = ntohs(qsr->port_no);
2106 queue_id = ntohl(qsr->queue_id);
2107 if (port_no == OFPP_ALL) {
2108 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2109 handle_queue_stats_for_port(port, queue_id, &cbdata);
2111 } else if (port_no < OFPP_MAX) {
2112 port = ofproto_get_port(ofproto, port_no);
2114 handle_queue_stats_for_port(port, queue_id, &cbdata);
2117 ofpbuf_list_delete(&cbdata.replies);
2118 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2120 ofconn_send_replies(ofconn, &cbdata.replies);
2126 is_flow_deletion_pending(const struct ofproto *ofproto,
2127 const struct cls_rule *cls_rule,
2130 if (!hmap_is_empty(&ofproto->deletions)) {
2131 struct ofoperation *op;
2133 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2134 cls_rule_hash(cls_rule, table_id),
2135 &ofproto->deletions) {
2136 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2145 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2146 * in which no matching flow already exists in the flow table.
2148 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2149 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2150 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2152 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2155 add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
2156 const struct ofp_header *request)
2158 struct classifier *table;
2159 struct ofopgroup *group;
2160 struct rule *victim;
2164 /* Check for overlap, if requested. */
2165 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2166 struct classifier *cls;
2168 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2169 if (classifier_rule_overlaps(cls, &fm->cr)) {
2170 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2176 if (fm->table_id == 0xff) {
2178 if (ofproto->n_tables > 1) {
2179 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2184 assert(table_id < ofproto->n_tables);
2185 table = &ofproto->tables[table_id];
2187 table = &ofproto->tables[0];
2189 } else if (fm->table_id < ofproto->n_tables) {
2190 table = &ofproto->tables[fm->table_id];
2192 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2195 /* Serialize against pending deletion. */
2196 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2197 return OFPROTO_POSTPONE;
2200 /* Allocate new rule. */
2201 rule = ofproto->ofproto_class->rule_alloc();
2203 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2204 ofproto->name, strerror(error));
2207 rule->ofproto = ofproto;
2209 rule->pending = NULL;
2210 rule->flow_cookie = fm->cookie;
2211 rule->created = time_msec();
2212 rule->idle_timeout = fm->idle_timeout;
2213 rule->hard_timeout = fm->hard_timeout;
2214 rule->table_id = table - ofproto->tables;
2215 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2216 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2217 rule->n_actions = fm->n_actions;
2219 /* Insert new rule. */
2220 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2221 if (victim && victim->pending) {
2222 error = OFPROTO_POSTPONE;
2225 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2226 : ofopgroup_create(ofproto));
2227 ofoperation_create(group, rule, OFOPERATION_ADD);
2228 rule->pending->victim = victim;
2230 error = ofproto->ofproto_class->rule_construct(rule);
2232 ofoperation_destroy(rule->pending);
2234 ofopgroup_submit(group);
2237 /* Back out if an error occurred. */
2240 classifier_replace(table, &victim->cr);
2242 classifier_remove(table, &rule->cr);
2244 ofproto_rule_destroy__(rule);
2249 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2251 /* Modifies the rules listed in 'rules', changing their actions to match those
2254 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2257 * Returns 0 on success, otherwise an OpenFlow error code. */
2259 modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
2260 const struct ofp_header *request, struct list *rules)
2262 struct ofopgroup *group;
2265 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2266 LIST_FOR_EACH (rule, ofproto_node, rules) {
2267 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2268 rule->actions, rule->n_actions)) {
2269 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2270 rule->pending->actions = rule->actions;
2271 rule->pending->n_actions = rule->n_actions;
2272 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2273 rule->n_actions = fm->n_actions;
2274 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2276 rule->flow_cookie = fm->cookie;
2278 ofopgroup_submit(group);
2283 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2284 * encoded by ofp_mkerr() on failure.
2286 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2289 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
2290 const struct ofp_header *request)
2292 struct ofproto *p = ofconn_get_ofproto(ofconn);
2296 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2297 return (error ? error
2298 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2299 : modify_flows__(ofconn, fm, request, &rules));
2302 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2303 * code as encoded by ofp_mkerr() on failure.
2305 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2308 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2309 const struct ofp_header *request)
2311 struct ofproto *p = ofconn_get_ofproto(ofconn);
2315 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2316 return (error ? error
2317 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2318 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2323 /* OFPFC_DELETE implementation. */
2325 /* Deletes the rules listed in 'rules'.
2327 * Returns 0 on success, otherwise an OpenFlow error code. */
2329 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2332 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2333 struct rule *rule, *next;
2334 struct ofopgroup *group;
2336 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2337 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2338 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2340 ofoperation_create(group, rule, OFOPERATION_DELETE);
2341 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2342 rule->ofproto->ofproto_class->rule_destruct(rule);
2344 ofopgroup_submit(group);
2349 /* Implements OFPFC_DELETE. */
2351 delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
2352 const struct ofp_header *request)
2354 struct ofproto *p = ofconn_get_ofproto(ofconn);
2358 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2360 return (error ? error
2361 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2365 /* Implements OFPFC_DELETE_STRICT. */
2367 delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2368 const struct ofp_header *request)
2370 struct ofproto *p = ofconn_get_ofproto(ofconn);
2374 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2376 return (error ? error
2377 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2383 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2385 struct ofputil_flow_removed fr;
2387 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2392 fr.cookie = rule->flow_cookie;
2394 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2395 fr.idle_timeout = rule->idle_timeout;
2396 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2399 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2402 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2403 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2406 * ofproto implementation ->run() functions should use this function to expire
2407 * OpenFlow flows. */
2409 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2411 struct ofproto *ofproto = rule->ofproto;
2412 struct ofopgroup *group;
2414 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2416 ofproto_rule_send_removed(rule, reason);
2418 group = ofopgroup_create(ofproto);
2419 ofoperation_create(group, rule, OFOPERATION_DELETE);
2420 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2421 rule->ofproto->ofproto_class->rule_destruct(rule);
2422 ofopgroup_submit(group);
2426 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2428 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2432 error = reject_slave_controller(ofconn, "flow_mod");
2437 if (list_size(&ofproto->pending) >= 50) {
2438 return OFPROTO_POSTPONE;
2441 error = ofputil_decode_flow_mod(&fm, oh,
2442 ofconn_get_flow_mod_table_id(ofconn));
2447 /* We do not support the emergency flow cache. It will hopefully get
2448 * dropped from OpenFlow in the near future. */
2449 if (fm.flags & OFPFF_EMERG) {
2450 /* There isn't a good fit for an error code, so just state that the
2451 * flow table is full. */
2452 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2455 switch (fm.command) {
2457 return add_flow(ofproto, ofconn, &fm, oh);
2460 return modify_flows_loose(ofconn, &fm, oh);
2462 case OFPFC_MODIFY_STRICT:
2463 return modify_flow_strict(ofconn, &fm, oh);
2466 return delete_flows_loose(ofconn, &fm, oh);
2468 case OFPFC_DELETE_STRICT:
2469 return delete_flow_strict(ofconn, &fm, oh);
2472 if (fm.command > 0xff) {
2473 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2474 "flow_mod_table_id extension is not enabled");
2476 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2481 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2483 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2484 struct nx_role_request *reply;
2488 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2489 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2490 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2493 role = ntohl(nrr->role);
2494 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2495 && role != NX_ROLE_SLAVE) {
2496 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2498 /* There's no good error code for this. */
2499 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2502 if (ofconn_get_role(ofconn) != role
2503 && ofconn_has_pending_opgroups(ofconn)) {
2504 return OFPROTO_POSTPONE;
2507 ofconn_set_role(ofconn, role);
2509 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2510 reply->role = htonl(role);
2511 ofconn_send_reply(ofconn, buf);
2517 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2518 const struct ofp_header *oh)
2520 const struct nxt_flow_mod_table_id *msg
2521 = (const struct nxt_flow_mod_table_id *) oh;
2523 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2528 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2530 const struct nxt_set_flow_format *msg
2531 = (const struct nxt_set_flow_format *) oh;
2534 format = ntohl(msg->format);
2535 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2536 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2539 if (format != ofconn_get_flow_format(ofconn)
2540 && ofconn_has_pending_opgroups(ofconn)) {
2541 /* Avoid sending async messages in surprising flow format. */
2542 return OFPROTO_POSTPONE;
2545 ofconn_set_flow_format(ofconn, format);
2550 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2552 struct ofp_header *ob;
2555 if (ofconn_has_pending_opgroups(ofconn)) {
2556 return OFPROTO_POSTPONE;
2559 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2560 ofconn_send_reply(ofconn, buf);
2565 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2567 const struct ofp_header *oh = msg->data;
2568 const struct ofputil_msg_type *type;
2571 error = ofputil_decode_msg_type(oh, &type);
2576 switch (ofputil_msg_type_code(type)) {
2577 /* OpenFlow requests. */
2578 case OFPUTIL_OFPT_ECHO_REQUEST:
2579 return handle_echo_request(ofconn, oh);
2581 case OFPUTIL_OFPT_FEATURES_REQUEST:
2582 return handle_features_request(ofconn, oh);
2584 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2585 return handle_get_config_request(ofconn, oh);
2587 case OFPUTIL_OFPT_SET_CONFIG:
2588 return handle_set_config(ofconn, msg->data);
2590 case OFPUTIL_OFPT_PACKET_OUT:
2591 return handle_packet_out(ofconn, oh);
2593 case OFPUTIL_OFPT_PORT_MOD:
2594 return handle_port_mod(ofconn, oh);
2596 case OFPUTIL_OFPT_FLOW_MOD:
2597 return handle_flow_mod(ofconn, oh);
2599 case OFPUTIL_OFPT_BARRIER_REQUEST:
2600 return handle_barrier_request(ofconn, oh);
2602 /* OpenFlow replies. */
2603 case OFPUTIL_OFPT_ECHO_REPLY:
2606 /* Nicira extension requests. */
2607 case OFPUTIL_NXT_ROLE_REQUEST:
2608 return handle_role_request(ofconn, oh);
2610 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2611 return handle_nxt_flow_mod_table_id(ofconn, oh);
2613 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2614 return handle_nxt_set_flow_format(ofconn, oh);
2616 case OFPUTIL_NXT_FLOW_MOD:
2617 return handle_flow_mod(ofconn, oh);
2619 /* Statistics requests. */
2620 case OFPUTIL_OFPST_DESC_REQUEST:
2621 return handle_desc_stats_request(ofconn, msg->data);
2623 case OFPUTIL_OFPST_FLOW_REQUEST:
2624 case OFPUTIL_NXST_FLOW_REQUEST:
2625 return handle_flow_stats_request(ofconn, msg->data);
2627 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2628 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2629 return handle_aggregate_stats_request(ofconn, msg->data);
2631 case OFPUTIL_OFPST_TABLE_REQUEST:
2632 return handle_table_stats_request(ofconn, msg->data);
2634 case OFPUTIL_OFPST_PORT_REQUEST:
2635 return handle_port_stats_request(ofconn, msg->data);
2637 case OFPUTIL_OFPST_QUEUE_REQUEST:
2638 return handle_queue_stats_request(ofconn, msg->data);
2640 case OFPUTIL_MSG_INVALID:
2641 case OFPUTIL_OFPT_HELLO:
2642 case OFPUTIL_OFPT_ERROR:
2643 case OFPUTIL_OFPT_FEATURES_REPLY:
2644 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2645 case OFPUTIL_OFPT_PACKET_IN:
2646 case OFPUTIL_OFPT_FLOW_REMOVED:
2647 case OFPUTIL_OFPT_PORT_STATUS:
2648 case OFPUTIL_OFPT_BARRIER_REPLY:
2649 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2650 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2651 case OFPUTIL_OFPST_DESC_REPLY:
2652 case OFPUTIL_OFPST_FLOW_REPLY:
2653 case OFPUTIL_OFPST_QUEUE_REPLY:
2654 case OFPUTIL_OFPST_PORT_REPLY:
2655 case OFPUTIL_OFPST_TABLE_REPLY:
2656 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2657 case OFPUTIL_NXT_ROLE_REPLY:
2658 case OFPUTIL_NXT_FLOW_REMOVED:
2659 case OFPUTIL_NXST_FLOW_REPLY:
2660 case OFPUTIL_NXST_AGGREGATE_REPLY:
2662 if (VLOG_IS_WARN_ENABLED()) {
2663 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2664 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2667 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2668 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2670 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2676 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2678 int error = handle_openflow__(ofconn, ofp_msg);
2679 if (error && error != OFPROTO_POSTPONE) {
2680 ofconn_send_error(ofconn, ofp_msg->data, error);
2682 COVERAGE_INC(ofproto_recv_openflow);
2683 return error != OFPROTO_POSTPONE;
2686 /* Asynchronous operations. */
2688 /* Creates and returns a new ofopgroup that is not associated with any
2689 * OpenFlow connection.
2691 * The caller should add operations to the returned group with
2692 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2693 static struct ofopgroup *
2694 ofopgroup_create(struct ofproto *ofproto)
2696 struct ofopgroup *group = xzalloc(sizeof *group);
2697 group->ofproto = ofproto;
2698 list_init(&group->ofproto_node);
2699 list_init(&group->ops);
2700 list_init(&group->ofconn_node);
2704 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2705 * the ofopgroup eventually fails, then the error reply will include 'request'.
2706 * If the ofopgroup eventually succeeds, then the packet with buffer id
2707 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2709 * The caller should add operations to the returned group with
2710 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2711 static struct ofopgroup *
2712 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2713 const struct ofp_header *request,
2716 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2717 size_t request_len = ntohs(request->length);
2719 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2720 group->ofconn = ofconn;
2721 group->request = xmemdup(request, MIN(request_len, 64));
2722 group->buffer_id = buffer_id;
2727 /* Submits 'group' for processing.
2729 * If 'group' contains no operations (e.g. none were ever added, or all of the
2730 * ones that were added completed synchronously), then it is destroyed
2731 * immediately. Otherwise it is added to the ofproto's list of pending
2734 ofopgroup_submit(struct ofopgroup *group)
2736 if (list_is_empty(&group->ops)) {
2737 ofopgroup_destroy(group);
2739 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2744 ofopgroup_destroy(struct ofopgroup *group)
2746 assert(list_is_empty(&group->ops));
2747 if (!list_is_empty(&group->ofproto_node)) {
2748 list_remove(&group->ofproto_node);
2750 if (!list_is_empty(&group->ofconn_node)) {
2751 list_remove(&group->ofconn_node);
2753 ofconn_send_error(group->ofconn, group->request, group->error);
2755 connmgr_retry(group->ofproto->connmgr);
2757 free(group->request);
2761 /* Initiates a new operation on 'rule', of the specified 'type', within
2762 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2764 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2765 enum ofoperation_type type)
2767 struct ofoperation *op;
2769 assert(!rule->pending);
2771 op = rule->pending = xzalloc(sizeof *op);
2773 list_push_back(&group->ops, &op->group_node);
2777 op->flow_cookie = rule->flow_cookie;
2779 if (type == OFOPERATION_DELETE) {
2780 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2781 cls_rule_hash(&rule->cr, rule->table_id));
2786 ofoperation_destroy(struct ofoperation *op)
2788 struct ofopgroup *group = op->group;
2791 op->rule->pending = NULL;
2793 if (op->type == OFOPERATION_DELETE) {
2794 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2796 list_remove(&op->group_node);
2800 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2801 ofopgroup_destroy(group);
2805 /* Indicates that 'op' completed with status 'error', which is either 0 to
2806 * indicate success or an OpenFlow error code (constructed with
2807 * e.g. ofp_mkerr()).
2809 * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
2810 * deletions are not allowed to fail.
2812 * Please see the large comment in ofproto/private.h titled "Asynchronous
2813 * Operation Support" for more information. */
2815 ofoperation_complete(struct ofoperation *op, int error)
2817 struct ofopgroup *group = op->group;
2818 struct rule *rule = op->rule;
2819 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2821 assert(rule->pending == op);
2822 assert(op->status < 0);
2827 && op->type != OFOPERATION_DELETE
2829 && group->buffer_id != UINT32_MAX
2830 && list_is_singleton(&op->group_node)) {
2831 struct ofpbuf *packet;
2834 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2838 error = rule_execute(rule, in_port, packet);
2841 if (!group->error) {
2842 group->error = error;
2846 case OFOPERATION_ADD:
2849 ofproto_rule_destroy__(op->victim);
2853 classifier_replace(table, &op->victim->cr);
2856 classifier_remove(table, &rule->cr);
2858 ofproto_rule_destroy__(rule);
2863 case OFOPERATION_DELETE:
2865 ofproto_rule_destroy__(rule);
2869 case OFOPERATION_MODIFY:
2871 free(rule->actions);
2872 rule->actions = op->actions;
2873 rule->n_actions = op->n_actions;
2881 ofoperation_destroy(op);
2885 ofoperation_get_victim(struct ofoperation *op)
2887 assert(op->type == OFOPERATION_ADD);
2892 pick_datapath_id(const struct ofproto *ofproto)
2894 const struct ofport *port;
2896 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2898 uint8_t ea[ETH_ADDR_LEN];
2901 error = netdev_get_etheraddr(port->netdev, ea);
2903 return eth_addr_to_uint64(ea);
2905 VLOG_WARN("could not get MAC address for %s (%s)",
2906 netdev_get_name(port->netdev), strerror(error));
2908 return ofproto->fallback_dpid;
2912 pick_fallback_dpid(void)
2914 uint8_t ea[ETH_ADDR_LEN];
2915 eth_addr_nicira_random(ea);
2916 return eth_addr_to_uint64(ea);
2919 /* unixctl commands. */
2922 ofproto_lookup(const char *name)
2924 struct ofproto *ofproto;
2926 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2928 if (!strcmp(ofproto->name, name)) {
2936 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2937 void *aux OVS_UNUSED)
2939 struct ofproto *ofproto;
2943 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2944 ds_put_format(&results, "%s\n", ofproto->name);
2946 unixctl_command_reply(conn, 200, ds_cstr(&results));
2947 ds_destroy(&results);
2951 ofproto_unixctl_init(void)
2953 static bool registered;
2959 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);