2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
24 #include "byte-order.h"
25 #include "classifier.h"
28 #include "dynamic-string.h"
33 #include "ofp-print.h"
36 #include "ofproto-provider.h"
37 #include "openflow/nicira-ext.h"
38 #include "openflow/openflow.h"
42 #include "poll-loop.h"
46 #include "unaligned.h"
50 VLOG_DEFINE_THIS_MODULE(ofproto);
52 COVERAGE_DEFINE(ofproto_error);
53 COVERAGE_DEFINE(ofproto_flush);
54 COVERAGE_DEFINE(ofproto_no_packet_in);
55 COVERAGE_DEFINE(ofproto_packet_out);
56 COVERAGE_DEFINE(ofproto_queue_req);
57 COVERAGE_DEFINE(ofproto_recv_openflow);
58 COVERAGE_DEFINE(ofproto_reinit_ports);
59 COVERAGE_DEFINE(ofproto_uninstallable);
60 COVERAGE_DEFINE(ofproto_update_port);
63 S_OPENFLOW, /* Processing OpenFlow commands. */
64 S_FLUSH, /* Deleting all flow table rules. */
67 enum ofoperation_type {
73 /* A single OpenFlow request can execute any number of operations. The
74 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
75 * ofconn to which an error reply should be sent if necessary.
77 * ofproto initiates some operations internally. These operations are still
78 * assigned to groups but will not have an associated ofconn. */
80 struct ofproto *ofproto; /* Owning ofproto. */
81 struct list ofproto_node; /* In ofproto's "pending" list. */
82 struct list ops; /* List of "struct ofoperation"s. */
84 /* Data needed to send OpenFlow reply on failure or to send a buffered
87 * If list_is_empty(ofconn_node) then this ofopgroup never had an
88 * associated ofconn or its ofconn's connection dropped after it initiated
89 * the operation. In the latter case 'ofconn' is a wild pointer that
90 * refers to freed memory, so the 'ofconn' member must be used only if
91 * !list_is_empty(ofconn_node).
93 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
94 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
95 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
96 uint32_t buffer_id; /* Buffer id from original request. */
97 int error; /* 0 if no error yet, otherwise error code. */
100 static struct ofopgroup *ofopgroup_create(struct ofproto *);
101 static struct ofopgroup *ofopgroup_create_for_ofconn(struct ofconn *,
102 const struct ofp_header *,
104 static void ofopgroup_submit(struct ofopgroup *);
105 static void ofopgroup_destroy(struct ofopgroup *);
107 /* A single flow table operation. */
109 struct ofopgroup *group; /* Owning group. */
110 struct list group_node; /* In ofopgroup's "ops" list. */
111 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
112 struct rule *rule; /* Rule being operated upon. */
113 enum ofoperation_type type; /* Type of operation. */
114 int status; /* -1 if pending, otherwise 0 or error code. */
115 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
116 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
117 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
118 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
121 static void ofoperation_create(struct ofopgroup *, struct rule *,
122 enum ofoperation_type);
123 static void ofoperation_destroy(struct ofoperation *);
125 static void ofport_destroy__(struct ofport *);
126 static void ofport_destroy(struct ofport *);
128 static uint64_t pick_datapath_id(const struct ofproto *);
129 static uint64_t pick_fallback_dpid(void);
131 static void ofproto_destroy__(struct ofproto *);
133 static void ofproto_rule_destroy__(struct rule *);
134 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
136 static void ofopgroup_destroy(struct ofopgroup *);
138 static int add_flow(struct ofproto *, struct ofconn *, struct flow_mod *,
139 const struct ofp_header *);
141 /* This return value tells handle_openflow() that processing of the current
142 * OpenFlow message must be postponed until some ongoing operations have
145 * This particular value is a good choice because it is negative (so it won't
146 * collide with any errno value or any value returned by ofp_mkerr()) and large
147 * (so it won't accidentally collide with EOF or a negative errno value). */
148 enum { OFPROTO_POSTPONE = -100000 };
150 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
152 static void update_port(struct ofproto *, const char *devname);
153 static int init_ports(struct ofproto *);
154 static void reinit_ports(struct ofproto *);
156 static void ofproto_unixctl_init(void);
158 /* All registered ofproto classes, in probe order. */
159 static const struct ofproto_class **ofproto_classes;
160 static size_t n_ofproto_classes;
161 static size_t allocated_ofproto_classes;
163 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
164 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
166 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
169 ofproto_initialize(void)
175 ofproto_class_register(&ofproto_dpif_class);
179 /* 'type' should be a normalized datapath type, as returned by
180 * ofproto_normalize_type(). Returns the corresponding ofproto_class
181 * structure, or a null pointer if there is none registered for 'type'. */
182 static const struct ofproto_class *
183 ofproto_class_find__(const char *type)
187 ofproto_initialize();
188 for (i = 0; i < n_ofproto_classes; i++) {
189 const struct ofproto_class *class = ofproto_classes[i];
194 class->enumerate_types(&types);
195 found = sset_contains(&types, type);
196 sset_destroy(&types);
202 VLOG_WARN("unknown datapath type %s", type);
206 /* Registers a new ofproto class. After successful registration, new ofprotos
207 * of that type can be created using ofproto_create(). */
209 ofproto_class_register(const struct ofproto_class *new_class)
213 for (i = 0; i < n_ofproto_classes; i++) {
214 if (ofproto_classes[i] == new_class) {
219 if (n_ofproto_classes >= allocated_ofproto_classes) {
220 ofproto_classes = x2nrealloc(ofproto_classes,
221 &allocated_ofproto_classes,
222 sizeof *ofproto_classes);
224 ofproto_classes[n_ofproto_classes++] = new_class;
228 /* Unregisters a datapath provider. 'type' must have been previously
229 * registered and not currently be in use by any ofprotos. After
230 * unregistration new datapaths of that type cannot be opened using
231 * ofproto_create(). */
233 ofproto_class_unregister(const struct ofproto_class *class)
237 for (i = 0; i < n_ofproto_classes; i++) {
238 if (ofproto_classes[i] == class) {
239 for (i++; i < n_ofproto_classes; i++) {
240 ofproto_classes[i - 1] = ofproto_classes[i];
246 VLOG_WARN("attempted to unregister an ofproto class that is not "
251 /* Clears 'types' and enumerates all registered ofproto types into it. The
252 * caller must first initialize the sset. */
254 ofproto_enumerate_types(struct sset *types)
258 ofproto_initialize();
259 for (i = 0; i < n_ofproto_classes; i++) {
260 ofproto_classes[i]->enumerate_types(types);
264 /* Returns the fully spelled out name for the given ofproto 'type'.
266 * Normalized type string can be compared with strcmp(). Unnormalized type
267 * string might be the same even if they have different spellings. */
269 ofproto_normalize_type(const char *type)
271 return type && type[0] ? type : "system";
274 /* Clears 'names' and enumerates the names of all known created ofprotos with
275 * the given 'type'. The caller must first initialize the sset. Returns 0 if
276 * successful, otherwise a positive errno value.
278 * Some kinds of datapaths might not be practically enumerable. This is not
279 * considered an error. */
281 ofproto_enumerate_names(const char *type, struct sset *names)
283 const struct ofproto_class *class = ofproto_class_find__(type);
284 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
288 ofproto_create(const char *datapath_name, const char *datapath_type,
289 struct ofproto **ofprotop)
291 const struct ofproto_class *class;
292 struct classifier *table;
293 struct ofproto *ofproto;
299 ofproto_initialize();
300 ofproto_unixctl_init();
302 datapath_type = ofproto_normalize_type(datapath_type);
303 class = ofproto_class_find__(datapath_type);
305 VLOG_WARN("could not create datapath %s of unknown type %s",
306 datapath_name, datapath_type);
310 ofproto = class->alloc();
312 VLOG_ERR("failed to allocate datapath %s of type %s",
313 datapath_name, datapath_type);
318 memset(ofproto, 0, sizeof *ofproto);
319 ofproto->ofproto_class = class;
320 ofproto->name = xstrdup(datapath_name);
321 ofproto->type = xstrdup(datapath_type);
322 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
323 hash_string(ofproto->name, 0));
324 ofproto->datapath_id = 0;
325 ofproto_set_flow_eviction_threshold(ofproto,
326 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
327 ofproto->forward_bpdu = false;
328 ofproto->fallback_dpid = pick_fallback_dpid();
329 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
330 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
331 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
332 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
333 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
334 hmap_init(&ofproto->ports);
335 shash_init(&ofproto->port_by_name);
336 ofproto->tables = NULL;
337 ofproto->n_tables = 0;
338 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
339 ofproto->state = S_OPENFLOW;
340 list_init(&ofproto->pending);
341 hmap_init(&ofproto->deletions);
343 error = ofproto->ofproto_class->construct(ofproto, &n_tables);
345 VLOG_ERR("failed to open datapath %s: %s",
346 datapath_name, strerror(error));
347 ofproto_destroy__(ofproto);
351 assert(n_tables >= 1 && n_tables <= 255);
352 ofproto->n_tables = n_tables;
353 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
354 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
355 classifier_init(table);
358 ofproto->datapath_id = pick_datapath_id(ofproto);
359 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
367 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
369 uint64_t old_dpid = p->datapath_id;
370 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
371 if (p->datapath_id != old_dpid) {
372 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
374 /* Force all active connections to reconnect, since there is no way to
375 * notify a controller that the datapath ID has changed. */
376 ofproto_reconnect_controllers(p);
381 ofproto_set_controllers(struct ofproto *p,
382 const struct ofproto_controller *controllers,
383 size_t n_controllers)
385 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
389 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
391 connmgr_set_fail_mode(p->connmgr, fail_mode);
394 /* Drops the connections between 'ofproto' and all of its controllers, forcing
395 * them to reconnect. */
397 ofproto_reconnect_controllers(struct ofproto *ofproto)
399 connmgr_reconnect(ofproto->connmgr);
402 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
403 * in-band control should guarantee access, in the same way that in-band
404 * control guarantees access to OpenFlow controllers. */
406 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
407 const struct sockaddr_in *extras, size_t n)
409 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
412 /* Sets the OpenFlow queue used by flows set up by in-band control on
413 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
414 * flows will use the default queue. */
416 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
418 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
421 /* Sets the number of flows at which eviction from the kernel flow table
424 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
426 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
427 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
429 ofproto->flow_eviction_threshold = threshold;
433 /* If forward_bpdu is true, the NORMAL action will forward frames with
434 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
435 * the NORMAL action will drop these frames. */
437 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
439 bool old_val = ofproto->forward_bpdu;
440 ofproto->forward_bpdu = forward_bpdu;
441 if (old_val != ofproto->forward_bpdu) {
442 if (ofproto->ofproto_class->forward_bpdu_changed) {
443 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
449 ofproto_set_desc(struct ofproto *p,
450 const char *mfr_desc, const char *hw_desc,
451 const char *sw_desc, const char *serial_desc,
454 struct ofp_desc_stats *ods;
457 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
458 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
459 sizeof ods->mfr_desc);
462 p->mfr_desc = xstrdup(mfr_desc);
465 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
466 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
467 sizeof ods->hw_desc);
470 p->hw_desc = xstrdup(hw_desc);
473 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
474 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
475 sizeof ods->sw_desc);
478 p->sw_desc = xstrdup(sw_desc);
481 if (strlen(serial_desc) >= sizeof ods->serial_num) {
482 VLOG_WARN("truncating serial_desc, must be less than %zu "
484 sizeof ods->serial_num);
486 free(p->serial_desc);
487 p->serial_desc = xstrdup(serial_desc);
490 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
491 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
492 sizeof ods->dp_desc);
495 p->dp_desc = xstrdup(dp_desc);
500 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
502 return connmgr_set_snoops(ofproto->connmgr, snoops);
506 ofproto_set_netflow(struct ofproto *ofproto,
507 const struct netflow_options *nf_options)
509 if (nf_options && sset_is_empty(&nf_options->collectors)) {
513 if (ofproto->ofproto_class->set_netflow) {
514 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
516 return nf_options ? EOPNOTSUPP : 0;
521 ofproto_set_sflow(struct ofproto *ofproto,
522 const struct ofproto_sflow_options *oso)
524 if (oso && sset_is_empty(&oso->targets)) {
528 if (ofproto->ofproto_class->set_sflow) {
529 return ofproto->ofproto_class->set_sflow(ofproto, oso);
531 return oso ? EOPNOTSUPP : 0;
535 /* Connectivity Fault Management configuration. */
537 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
539 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
541 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
542 if (ofport && ofproto->ofproto_class->set_cfm) {
543 ofproto->ofproto_class->set_cfm(ofport, NULL);
547 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
548 * basic configuration from the configuration members in 'cfm', and the remote
549 * maintenance point ID from remote_mpid. Ignores the statistics members of
552 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
554 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
555 const struct cfm_settings *s)
557 struct ofport *ofport;
560 ofport = ofproto_get_port(ofproto, ofp_port);
562 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
563 ofproto->name, ofp_port);
567 /* XXX: For configuration simplicity, we only support one remote_mpid
568 * outside of the CFM module. It's not clear if this is the correct long
569 * term solution or not. */
570 error = (ofproto->ofproto_class->set_cfm
571 ? ofproto->ofproto_class->set_cfm(ofport, s)
574 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
575 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
580 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
581 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
582 * 0 if LACP partner information is not current (generally indicating a
583 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
585 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
587 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
588 return (ofport && ofproto->ofproto_class->port_is_lacp_current
589 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
595 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
596 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
597 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
598 * configuration plus, if there is more than one slave, a bonding
601 * If 'aux' is already registered then this function updates its configuration
602 * to 's'. Otherwise, this function registers a new bundle.
604 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
607 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
608 const struct ofproto_bundle_settings *s)
610 return (ofproto->ofproto_class->bundle_set
611 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
615 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
616 * If no such bundle has been registered, this has no effect. */
618 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
620 return ofproto_bundle_register(ofproto, aux, NULL);
624 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
625 * If 'aux' is already registered then this function updates its configuration
626 * to 's'. Otherwise, this function registers a new mirror.
628 * Mirrors affect only the treatment of packets output to the OFPP_NORMAL
631 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
632 const struct ofproto_mirror_settings *s)
634 return (ofproto->ofproto_class->mirror_set
635 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
639 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
640 * If no mirror has been registered, this has no effect. */
642 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
644 return ofproto_mirror_register(ofproto, aux, NULL);
647 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
648 * which all packets are flooded, instead of using MAC learning. If
649 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
651 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
654 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
656 return (ofproto->ofproto_class->set_flood_vlans
657 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
661 /* Returns true if 'aux' is a registered bundle that is currently in use as the
662 * output for a mirror. */
664 ofproto_is_mirror_output_bundle(struct ofproto *ofproto, void *aux)
666 return (ofproto->ofproto_class->is_mirror_output_bundle
667 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
672 ofproto_has_snoops(const struct ofproto *ofproto)
674 return connmgr_has_snoops(ofproto->connmgr);
678 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
680 connmgr_get_snoops(ofproto->connmgr, snoops);
684 ofproto_flush__(struct ofproto *ofproto)
686 struct classifier *table;
687 struct ofopgroup *group;
689 if (ofproto->ofproto_class->flush) {
690 ofproto->ofproto_class->flush(ofproto);
693 group = ofopgroup_create(ofproto);
694 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
695 struct rule *rule, *next_rule;
696 struct cls_cursor cursor;
698 cls_cursor_init(&cursor, table, NULL);
699 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
700 if (!rule->pending) {
701 ofoperation_create(group, rule, OFOPERATION_DELETE);
702 classifier_remove(table, &rule->cr);
703 ofproto->ofproto_class->rule_destruct(rule);
707 ofopgroup_submit(group);
711 ofproto_destroy__(struct ofproto *ofproto)
713 struct classifier *table;
715 assert(list_is_empty(&ofproto->pending));
717 connmgr_destroy(ofproto->connmgr);
719 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
722 free(ofproto->mfr_desc);
723 free(ofproto->hw_desc);
724 free(ofproto->sw_desc);
725 free(ofproto->serial_desc);
726 free(ofproto->dp_desc);
727 hmap_destroy(&ofproto->ports);
728 shash_destroy(&ofproto->port_by_name);
730 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
731 assert(classifier_is_empty(table));
732 classifier_destroy(table);
734 free(ofproto->tables);
736 hmap_destroy(&ofproto->deletions);
738 ofproto->ofproto_class->dealloc(ofproto);
742 ofproto_destroy(struct ofproto *p)
744 struct ofport *ofport, *next_ofport;
751 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
752 ofport_destroy(ofport);
755 p->ofproto_class->destruct(p);
756 ofproto_destroy__(p);
759 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
760 * kernel datapath, for example, this destroys the datapath in the kernel, and
761 * with the netdev-based datapath, it tears down the data structures that
762 * represent the datapath.
764 * The datapath should not be currently open as an ofproto. */
766 ofproto_delete(const char *name, const char *type)
768 const struct ofproto_class *class = ofproto_class_find__(type);
769 return (!class ? EAFNOSUPPORT
770 : !class->del ? EACCES
771 : class->del(type, name));
775 process_port_change(struct ofproto *ofproto, int error, char *devname)
777 if (error == ENOBUFS) {
778 reinit_ports(ofproto);
780 update_port(ofproto, devname);
786 ofproto_run(struct ofproto *p)
788 struct ofport *ofport;
792 error = p->ofproto_class->run(p);
793 if (error == ENODEV) {
794 /* Someone destroyed the datapath behind our back. The caller
795 * better destroy us and give up, because we're just going to
796 * spin from here on out. */
797 static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5);
798 VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally",
803 if (p->ofproto_class->port_poll) {
804 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
805 process_port_change(p, error, devname);
809 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
810 unsigned int change_seq = netdev_change_seq(ofport->netdev);
811 if (ofport->change_seq != change_seq) {
812 ofport->change_seq = change_seq;
813 update_port(p, netdev_get_name(ofport->netdev));
820 connmgr_run(p->connmgr, handle_openflow);
824 connmgr_run(p->connmgr, NULL);
826 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
827 connmgr_flushed(p->connmgr);
828 p->state = S_OPENFLOW;
840 ofproto_wait(struct ofproto *p)
842 struct ofport *ofport;
844 p->ofproto_class->wait(p);
845 if (p->ofproto_class->port_poll_wait) {
846 p->ofproto_class->port_poll_wait(p);
849 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
850 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
851 poll_immediate_wake();
857 connmgr_wait(p->connmgr, true);
861 connmgr_wait(p->connmgr, false);
862 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
863 poll_immediate_wake();
870 ofproto_is_alive(const struct ofproto *p)
872 return connmgr_has_controllers(p->connmgr);
876 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
879 connmgr_get_controller_info(ofproto->connmgr, info);
883 ofproto_free_ofproto_controller_info(struct shash *info)
885 connmgr_free_controller_info(info);
888 /* Makes a deep copy of 'old' into 'port'. */
890 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
892 port->name = xstrdup(old->name);
893 port->type = xstrdup(old->type);
894 port->ofp_port = old->ofp_port;
897 /* Frees memory allocated to members of 'ofproto_port'.
899 * Do not call this function on an ofproto_port obtained from
900 * ofproto_port_dump_next(): that function retains ownership of the data in the
903 ofproto_port_destroy(struct ofproto_port *ofproto_port)
905 free(ofproto_port->name);
906 free(ofproto_port->type);
909 /* Initializes 'dump' to begin dumping the ports in an ofproto.
911 * This function provides no status indication. An error status for the entire
912 * dump operation is provided when it is completed by calling
913 * ofproto_port_dump_done().
916 ofproto_port_dump_start(struct ofproto_port_dump *dump,
917 const struct ofproto *ofproto)
919 dump->ofproto = ofproto;
920 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
924 /* Attempts to retrieve another port from 'dump', which must have been created
925 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
926 * 'port' and returns true. On failure, returns false.
928 * Failure might indicate an actual error or merely that the last port has been
929 * dumped. An error status for the entire dump operation is provided when it
930 * is completed by calling ofproto_port_dump_done().
932 * The ofproto owns the data stored in 'port'. It will remain valid until at
933 * least the next time 'dump' is passed to ofproto_port_dump_next() or
934 * ofproto_port_dump_done(). */
936 ofproto_port_dump_next(struct ofproto_port_dump *dump,
937 struct ofproto_port *port)
939 const struct ofproto *ofproto = dump->ofproto;
945 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
948 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
954 /* Completes port table dump operation 'dump', which must have been created
955 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
956 * error-free, otherwise a positive errno value describing the problem. */
958 ofproto_port_dump_done(struct ofproto_port_dump *dump)
960 const struct ofproto *ofproto = dump->ofproto;
962 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
965 return dump->error == EOF ? 0 : dump->error;
968 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
969 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
970 * is non-null). On failure, returns a positive errno value and sets
971 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
973 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
979 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
981 update_port(ofproto, netdev_get_name(netdev));
984 *ofp_portp = error ? OFPP_NONE : ofp_port;
989 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
990 * initializes '*port' appropriately; on failure, returns a positive errno
993 * The caller owns the data in 'ofproto_port' and must free it with
994 * ofproto_port_destroy() when it is no longer needed. */
996 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
997 struct ofproto_port *port)
1001 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1003 memset(port, 0, sizeof *port);
1008 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1009 * Returns 0 if successful, otherwise a positive errno. */
1011 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1013 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1014 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1017 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1018 if (!error && ofport) {
1019 /* 'name' is the netdev's name and update_port() is going to close the
1020 * netdev. Just in case update_port() refers to 'name' after it
1021 * destroys 'ofport', make a copy of it around the update_port()
1023 char *devname = xstrdup(name);
1024 update_port(ofproto, devname);
1030 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1031 * performs the 'n_actions' actions in 'actions'. The new flow will not
1034 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1035 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1036 * controllers; otherwise, it will be hidden.
1038 * The caller retains ownership of 'cls_rule' and 'actions'.
1040 * This is a helper function for in-band control and fail-open. */
1042 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1043 const union ofp_action *actions, size_t n_actions)
1045 const struct rule *rule;
1047 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1048 &ofproto->tables[0], cls_rule));
1049 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1050 actions, n_actions)) {
1053 memset(&fm, 0, sizeof fm);
1055 fm.buffer_id = UINT32_MAX;
1056 fm.actions = (union ofp_action *) actions;
1057 fm.n_actions = n_actions;
1058 add_flow(ofproto, NULL, &fm, NULL);
1062 /* Searches for a rule with matching criteria exactly equal to 'target' in
1063 * ofproto's table 0 and, if it finds one, deletes it.
1065 * This is a helper function for in-band control and fail-open. */
1067 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1071 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1072 &ofproto->tables[0], target));
1074 /* No such rule -> success. */
1076 } else if (rule->pending) {
1077 /* An operation on the rule is already pending -> failure.
1078 * Caller must retry later if it's important. */
1081 /* Initiate deletion -> success. */
1082 struct ofopgroup *group = ofopgroup_create(ofproto);
1083 ofoperation_create(group, rule, OFOPERATION_DELETE);
1084 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1085 rule->ofproto->ofproto_class->rule_destruct(rule);
1086 ofopgroup_submit(group);
1092 /* Starts the process of deleting all of the flows from all of ofproto's flow
1093 * tables and then reintroducing the flows required by in-band control and
1094 * fail-open. The process will complete in a later call to ofproto_run(). */
1096 ofproto_flush_flows(struct ofproto *ofproto)
1098 COVERAGE_INC(ofproto_flush);
1099 ofproto->state = S_FLUSH;
1103 reinit_ports(struct ofproto *p)
1105 struct ofproto_port_dump dump;
1106 struct sset devnames;
1107 struct ofport *ofport;
1108 struct ofproto_port ofproto_port;
1109 const char *devname;
1111 COVERAGE_INC(ofproto_reinit_ports);
1113 sset_init(&devnames);
1114 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1115 sset_add(&devnames, netdev_get_name(ofport->netdev));
1117 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1118 sset_add(&devnames, ofproto_port.name);
1121 SSET_FOR_EACH (devname, &devnames) {
1122 update_port(p, devname);
1124 sset_destroy(&devnames);
1127 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1128 * netdev cannot be opened. On success, also fills in 'opp'. */
1129 static struct netdev *
1130 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1132 uint32_t curr, advertised, supported, peer;
1133 enum netdev_flags flags;
1134 struct netdev *netdev;
1137 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1139 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1140 "cannot be opened (%s)",
1141 ofproto_port->name, ofproto_port->ofp_port,
1142 ofproto_port->name, strerror(error));
1146 netdev_get_flags(netdev, &flags);
1147 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1149 opp->port_no = htons(ofproto_port->ofp_port);
1150 netdev_get_etheraddr(netdev, opp->hw_addr);
1151 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1152 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1153 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1154 opp->curr = htonl(curr);
1155 opp->advertised = htonl(advertised);
1156 opp->supported = htonl(supported);
1157 opp->peer = htonl(peer);
1162 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1163 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1166 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1168 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1169 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1170 && a->state == b->state
1171 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1172 && a->curr == b->curr
1173 && a->advertised == b->advertised
1174 && a->supported == b->supported
1175 && a->peer == b->peer);
1178 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1179 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1180 * one with the same name or port number). */
1182 ofport_install(struct ofproto *p,
1183 struct netdev *netdev, const struct ofp_phy_port *opp)
1185 const char *netdev_name = netdev_get_name(netdev);
1186 struct ofport *ofport;
1189 /* Create ofport. */
1190 ofport = p->ofproto_class->port_alloc();
1195 ofport->ofproto = p;
1196 ofport->netdev = netdev;
1197 ofport->change_seq = netdev_change_seq(netdev);
1199 ofport->ofp_port = ntohs(opp->port_no);
1201 /* Add port to 'p'. */
1202 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1203 shash_add(&p->port_by_name, netdev_name, ofport);
1205 /* Let the ofproto_class initialize its private data. */
1206 error = p->ofproto_class->port_construct(ofport);
1210 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1214 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1215 p->name, netdev_name, strerror(error));
1217 ofport_destroy__(ofport);
1219 netdev_close(netdev);
1223 /* Removes 'ofport' from 'p' and destroys it. */
1225 ofport_remove(struct ofport *ofport)
1227 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1229 ofport_destroy(ofport);
1232 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1235 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1237 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1239 ofport_remove(port);
1243 /* Updates 'port' within 'ofproto' with the new 'netdev' and 'opp'.
1245 * Does not handle a name or port number change. The caller must implement
1246 * such a change as a delete followed by an add. */
1248 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1250 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1251 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1252 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1253 port->opp.state = opp->state;
1254 port->opp.curr = opp->curr;
1255 port->opp.advertised = opp->advertised;
1256 port->opp.supported = opp->supported;
1257 port->opp.peer = opp->peer;
1259 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1263 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1265 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1267 if (port->ofproto->ofproto_class->set_cfm) {
1268 port->ofproto->ofproto_class->set_cfm(port, NULL);
1270 if (port->ofproto->ofproto_class->bundle_remove) {
1271 port->ofproto->ofproto_class->bundle_remove(port);
1277 ofport_destroy__(struct ofport *port)
1279 struct ofproto *ofproto = port->ofproto;
1280 const char *name = netdev_get_name(port->netdev);
1282 hmap_remove(&ofproto->ports, &port->hmap_node);
1283 shash_delete(&ofproto->port_by_name,
1284 shash_find(&ofproto->port_by_name, name));
1286 netdev_close(port->netdev);
1287 ofproto->ofproto_class->port_dealloc(port);
1291 ofport_destroy(struct ofport *port)
1294 port->ofproto->ofproto_class->port_destruct(port);
1295 ofport_destroy__(port);
1300 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1302 struct ofport *port;
1304 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1305 hash_int(ofp_port, 0), &ofproto->ports) {
1306 if (port->ofp_port == ofp_port) {
1314 update_port(struct ofproto *ofproto, const char *name)
1316 struct ofproto_port ofproto_port;
1317 struct ofp_phy_port opp;
1318 struct netdev *netdev;
1319 struct ofport *port;
1321 COVERAGE_INC(ofproto_update_port);
1323 /* Fetch 'name''s location and properties from the datapath. */
1324 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1325 ? ofport_open(&ofproto_port, &opp)
1328 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1329 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1330 struct netdev *old_netdev = port->netdev;
1332 /* 'name' hasn't changed location. Any properties changed? */
1333 if (!ofport_equal(&port->opp, &opp)) {
1334 ofport_modified(port, &opp);
1337 /* Install the newly opened netdev in case it has changed.
1338 * Don't close the old netdev yet in case port_modified has to
1339 * remove a retained reference to it.*/
1340 port->netdev = netdev;
1341 port->change_seq = netdev_change_seq(netdev);
1343 if (port->ofproto->ofproto_class->port_modified) {
1344 port->ofproto->ofproto_class->port_modified(port);
1347 netdev_close(old_netdev);
1349 /* If 'port' is nonnull then its name differs from 'name' and thus
1350 * we should delete it. If we think there's a port named 'name'
1351 * then its port number must be wrong now so delete it too. */
1353 ofport_remove(port);
1355 ofport_remove_with_name(ofproto, name);
1356 ofport_install(ofproto, netdev, &opp);
1359 /* Any port named 'name' is gone now. */
1360 ofport_remove_with_name(ofproto, name);
1362 ofproto_port_destroy(&ofproto_port);
1366 init_ports(struct ofproto *p)
1368 struct ofproto_port_dump dump;
1369 struct ofproto_port ofproto_port;
1371 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1372 uint16_t ofp_port = ofproto_port.ofp_port;
1373 if (ofproto_get_port(p, ofp_port)) {
1374 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1376 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1377 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1380 struct ofp_phy_port opp;
1381 struct netdev *netdev;
1383 netdev = ofport_open(&ofproto_port, &opp);
1385 ofport_install(p, netdev, &opp);
1394 ofproto_rule_destroy__(struct rule *rule)
1396 free(rule->actions);
1397 rule->ofproto->ofproto_class->rule_dealloc(rule);
1400 /* This function allows an ofproto implementation to destroy any rules that
1401 * remain when its ->destruct() function is called. The caller must have
1402 * already uninitialized any derived members of 'rule' (step 5 described in the
1403 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1404 * This function implements steps 6 and 7.
1406 * This function should only be called from an ofproto implementation's
1407 * ->destruct() function. It is not suitable elsewhere. */
1409 ofproto_rule_destroy(struct rule *rule)
1411 assert(!rule->pending);
1412 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1413 ofproto_rule_destroy__(rule);
1416 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1417 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1420 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1422 const union ofp_action *oa;
1425 if (out_port == OFPP_NONE) {
1428 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1429 if (action_outputs_to_port(oa, htons(out_port))) {
1436 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1437 * statistics appropriately. 'packet' must have at least sizeof(struct
1438 * ofp_packet_in) bytes of headroom.
1440 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1441 * with statistics for 'packet' either way.
1443 * Takes ownership of 'packet'. */
1445 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1449 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1451 flow_extract(packet, 0, in_port, &flow);
1452 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1455 /* Returns true if 'rule' should be hidden from the controller.
1457 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1458 * (e.g. by in-band control) and are intentionally hidden from the
1461 rule_is_hidden(const struct rule *rule)
1463 return rule->cr.priority > UINT16_MAX;
1467 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1469 ofconn_send_reply(ofconn, make_echo_reply(oh));
1474 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1476 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1477 struct ofp_switch_features *osf;
1479 struct ofport *port;
1483 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1484 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1486 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1487 osf->datapath_id = htonll(ofproto->datapath_id);
1488 osf->n_buffers = htonl(pktbuf_capacity());
1489 osf->n_tables = ofproto->n_tables;
1490 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1493 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1495 osf->actions = htonl(actions);
1497 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1498 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1501 ofconn_send_reply(ofconn, buf);
1506 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1508 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1510 struct ofp_switch_config *osc;
1514 /* Figure out flags. */
1515 drop_frags = ofproto->ofproto_class->get_drop_frags(ofproto);
1516 flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL;
1519 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1520 osc->flags = htons(flags);
1521 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1522 ofconn_send_reply(ofconn, buf);
1528 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1530 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1531 uint16_t flags = ntohs(osc->flags);
1533 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1534 && ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1535 switch (flags & OFPC_FRAG_MASK) {
1536 case OFPC_FRAG_NORMAL:
1537 ofproto->ofproto_class->set_drop_frags(ofproto, false);
1539 case OFPC_FRAG_DROP:
1540 ofproto->ofproto_class->set_drop_frags(ofproto, true);
1543 VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")",
1549 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1554 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1555 * error message code (composed with ofp_mkerr()) for the caller to propagate
1556 * upward. Otherwise, returns 0.
1558 * The log message mentions 'msg_type'. */
1560 reject_slave_controller(struct ofconn *ofconn, const char *msg_type)
1562 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1563 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1564 static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5);
1565 VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller",
1568 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1575 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1577 struct ofproto *p = ofconn_get_ofproto(ofconn);
1578 struct ofp_packet_out *opo;
1579 struct ofpbuf payload, *buffer;
1580 union ofp_action *ofp_actions;
1581 struct ofpbuf request;
1583 size_t n_ofp_actions;
1587 COVERAGE_INC(ofproto_packet_out);
1589 error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT");
1594 /* Get ofp_packet_out. */
1595 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1596 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1599 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1600 &ofp_actions, &n_ofp_actions);
1606 if (opo->buffer_id != htonl(UINT32_MAX)) {
1607 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1609 if (error || !buffer) {
1618 /* Send out packet. */
1619 flow_extract(&payload, 0, ntohs(opo->in_port), &flow);
1620 error = p->ofproto_class->packet_out(p, &payload, &flow,
1621 ofp_actions, n_ofp_actions);
1622 ofpbuf_delete(buffer);
1628 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1630 ovs_be32 old_config = port->opp.config;
1632 mask &= config ^ port->opp.config;
1633 if (mask & htonl(OFPPC_PORT_DOWN)) {
1634 if (config & htonl(OFPPC_PORT_DOWN)) {
1635 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1637 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1641 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1642 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1643 OFPPC_NO_PACKET_IN));
1644 if (port->opp.config != old_config) {
1645 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1650 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1652 struct ofproto *p = ofconn_get_ofproto(ofconn);
1653 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1654 struct ofport *port;
1657 error = reject_slave_controller(ofconn, "OFPT_PORT_MOD");
1662 port = ofproto_get_port(p, ntohs(opm->port_no));
1664 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1665 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1666 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1668 update_port_config(port, opm->config, opm->mask);
1669 if (opm->advertise) {
1670 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1677 handle_desc_stats_request(struct ofconn *ofconn,
1678 const struct ofp_stats_msg *request)
1680 struct ofproto *p = ofconn_get_ofproto(ofconn);
1681 struct ofp_desc_stats *ods;
1684 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1685 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1686 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1687 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1688 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1689 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1690 ofconn_send_reply(ofconn, msg);
1696 handle_table_stats_request(struct ofconn *ofconn,
1697 const struct ofp_stats_msg *request)
1699 struct ofproto *p = ofconn_get_ofproto(ofconn);
1700 struct ofp_table_stats *ots;
1704 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1706 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1707 for (i = 0; i < p->n_tables; i++) {
1708 ots[i].table_id = i;
1709 sprintf(ots[i].name, "table%zu", i);
1710 ots[i].wildcards = htonl(OFPFW_ALL);
1711 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1712 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1715 p->ofproto_class->get_tables(p, ots);
1717 ofconn_send_reply(ofconn, msg);
1722 append_port_stat(struct ofport *port, struct list *replies)
1724 struct netdev_stats stats;
1725 struct ofp_port_stats *ops;
1727 /* Intentionally ignore return value, since errors will set
1728 * 'stats' to all-1s, which is correct for OpenFlow, and
1729 * netdev_get_stats() will log errors. */
1730 netdev_get_stats(port->netdev, &stats);
1732 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1733 ops->port_no = port->opp.port_no;
1734 memset(ops->pad, 0, sizeof ops->pad);
1735 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1736 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1737 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1738 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1739 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1740 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1741 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1742 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1743 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1744 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1745 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1746 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1750 handle_port_stats_request(struct ofconn *ofconn,
1751 const struct ofp_port_stats_request *psr)
1753 struct ofproto *p = ofconn_get_ofproto(ofconn);
1754 struct ofport *port;
1755 struct list replies;
1757 ofputil_start_stats_reply(&psr->osm, &replies);
1758 if (psr->port_no != htons(OFPP_NONE)) {
1759 port = ofproto_get_port(p, ntohs(psr->port_no));
1761 append_port_stat(port, &replies);
1764 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
1765 append_port_stat(port, &replies);
1769 ofconn_send_replies(ofconn, &replies);
1774 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
1776 long long int msecs = time_msec() - start;
1777 *sec = msecs / 1000;
1778 *nsec = (msecs % 1000) * (1000 * 1000);
1781 static struct classifier *
1782 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
1784 if (table_id == 0xff) {
1785 return &ofproto->tables[0];
1786 } else if (table_id < ofproto->n_tables) {
1787 return &ofproto->tables[table_id];
1789 /* It would probably be better to reply with an error but there doesn't
1790 * seem to be any appropriate value, so that might just be
1792 VLOG_WARN_RL(&rl, "controller asked for invalid table %"PRIu8,
1798 static struct classifier *
1799 next_matching_table(struct ofproto *ofproto,
1800 struct classifier *cls, uint8_t table_id)
1802 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
1807 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
1810 * - If TABLE_ID is 0xff, this iterates over every classifier table in
1813 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
1814 * only once, for that table.
1816 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so ofproto logs a warning
1817 * and does not enter the loop at all.
1819 * All parameters are evaluated multiple times.
1821 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
1822 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
1824 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
1826 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1827 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
1828 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
1831 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1832 * to 'out_port' are included.
1834 * Hidden rules are always omitted.
1836 * Returns 0 on success, otherwise an OpenFlow error code. */
1838 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
1839 const struct cls_rule *match, uint16_t out_port,
1842 struct classifier *cls;
1845 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1846 struct cls_cursor cursor;
1849 cls_cursor_init(&cursor, cls, match);
1850 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1851 if (rule->pending) {
1852 return OFPROTO_POSTPONE;
1854 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1855 list_push_back(rules, &rule->ofproto_node);
1862 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
1863 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
1864 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
1867 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
1868 * to 'out_port' are included.
1870 * Hidden rules are always omitted.
1872 * Returns 0 on success, otherwise an OpenFlow error code. */
1874 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
1875 const struct cls_rule *match, uint16_t out_port,
1878 struct classifier *cls;
1881 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
1884 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
1886 if (rule->pending) {
1887 return OFPROTO_POSTPONE;
1889 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
1890 list_push_back(rules, &rule->ofproto_node);
1898 handle_flow_stats_request(struct ofconn *ofconn,
1899 const struct ofp_stats_msg *osm)
1901 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1902 struct flow_stats_request fsr;
1903 struct list replies;
1908 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
1913 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
1914 fsr.out_port, &rules);
1919 ofputil_start_stats_reply(osm, &replies);
1920 LIST_FOR_EACH (rule, ofproto_node, &rules) {
1921 struct ofputil_flow_stats fs;
1924 fs.cookie = rule->flow_cookie;
1925 fs.table_id = rule->table_id;
1926 calc_flow_duration__(rule->created, &fs.duration_sec,
1928 fs.idle_timeout = rule->idle_timeout;
1929 fs.hard_timeout = rule->hard_timeout;
1930 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
1932 fs.actions = rule->actions;
1933 fs.n_actions = rule->n_actions;
1934 ofputil_append_flow_stats_reply(&fs, &replies);
1936 ofconn_send_replies(ofconn, &replies);
1942 flow_stats_ds(struct rule *rule, struct ds *results)
1944 uint64_t packet_count, byte_count;
1946 rule->ofproto->ofproto_class->rule_get_stats(rule,
1947 &packet_count, &byte_count);
1949 if (rule->table_id != 0) {
1950 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
1952 ds_put_format(results, "duration=%llds, ",
1953 (time_msec() - rule->created) / 1000);
1954 ds_put_format(results, "priority=%u, ", rule->cr.priority);
1955 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
1956 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
1957 cls_rule_format(&rule->cr, results);
1958 ds_put_char(results, ',');
1959 if (rule->n_actions > 0) {
1960 ofp_print_actions(results, rule->actions, rule->n_actions);
1962 ds_put_cstr(results, "drop");
1964 ds_put_cstr(results, "\n");
1967 /* Adds a pretty-printed description of all flows to 'results', including
1968 * hidden flows (e.g., set up by in-band control). */
1970 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
1972 struct classifier *cls;
1974 OFPROTO_FOR_EACH_TABLE (cls, p) {
1975 struct cls_cursor cursor;
1978 cls_cursor_init(&cursor, cls, NULL);
1979 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1980 flow_stats_ds(rule, results);
1985 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
1986 * '*engine_type' and '*engine_id', respectively. */
1988 ofproto_get_netflow_ids(const struct ofproto *ofproto,
1989 uint8_t *engine_type, uint8_t *engine_id)
1991 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
1994 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
1995 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
1996 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
1998 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2000 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2001 return (ofport && ofproto->ofproto_class->get_cfm_fault
2002 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2007 handle_aggregate_stats_request(struct ofconn *ofconn,
2008 const struct ofp_stats_msg *osm)
2010 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2011 struct flow_stats_request request;
2012 struct ofputil_aggregate_stats stats;
2013 bool unknown_packets, unknown_bytes;
2014 struct ofpbuf *reply;
2019 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2024 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2025 request.out_port, &rules);
2030 memset(&stats, 0, sizeof stats);
2031 unknown_packets = unknown_bytes = false;
2032 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2033 uint64_t packet_count;
2034 uint64_t byte_count;
2036 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2039 if (packet_count == UINT64_MAX) {
2040 unknown_packets = true;
2042 stats.packet_count += packet_count;
2045 if (byte_count == UINT64_MAX) {
2046 unknown_bytes = true;
2048 stats.byte_count += byte_count;
2053 if (unknown_packets) {
2054 stats.packet_count = UINT64_MAX;
2056 if (unknown_bytes) {
2057 stats.byte_count = UINT64_MAX;
2060 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2061 ofconn_send_reply(ofconn, reply);
2066 struct queue_stats_cbdata {
2067 struct ofport *ofport;
2068 struct list replies;
2072 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2073 const struct netdev_queue_stats *stats)
2075 struct ofp_queue_stats *reply;
2077 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2078 reply->port_no = cbdata->ofport->opp.port_no;
2079 memset(reply->pad, 0, sizeof reply->pad);
2080 reply->queue_id = htonl(queue_id);
2081 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2082 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2083 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2087 handle_queue_stats_dump_cb(uint32_t queue_id,
2088 struct netdev_queue_stats *stats,
2091 struct queue_stats_cbdata *cbdata = cbdata_;
2093 put_queue_stats(cbdata, queue_id, stats);
2097 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2098 struct queue_stats_cbdata *cbdata)
2100 cbdata->ofport = port;
2101 if (queue_id == OFPQ_ALL) {
2102 netdev_dump_queue_stats(port->netdev,
2103 handle_queue_stats_dump_cb, cbdata);
2105 struct netdev_queue_stats stats;
2107 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2108 put_queue_stats(cbdata, queue_id, &stats);
2114 handle_queue_stats_request(struct ofconn *ofconn,
2115 const struct ofp_queue_stats_request *qsr)
2117 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2118 struct queue_stats_cbdata cbdata;
2119 struct ofport *port;
2120 unsigned int port_no;
2123 COVERAGE_INC(ofproto_queue_req);
2125 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2127 port_no = ntohs(qsr->port_no);
2128 queue_id = ntohl(qsr->queue_id);
2129 if (port_no == OFPP_ALL) {
2130 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2131 handle_queue_stats_for_port(port, queue_id, &cbdata);
2133 } else if (port_no < OFPP_MAX) {
2134 port = ofproto_get_port(ofproto, port_no);
2136 handle_queue_stats_for_port(port, queue_id, &cbdata);
2139 ofpbuf_list_delete(&cbdata.replies);
2140 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2142 ofconn_send_replies(ofconn, &cbdata.replies);
2148 is_flow_deletion_pending(const struct ofproto *ofproto,
2149 const struct cls_rule *cls_rule,
2152 if (!hmap_is_empty(&ofproto->deletions)) {
2153 struct ofoperation *op;
2155 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2156 cls_rule_hash(cls_rule, table_id),
2157 &ofproto->deletions) {
2158 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2167 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2168 * in which no matching flow already exists in the flow table.
2170 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2171 * ofp_actions, to the ofproto's flow table. Returns 0 on success or an
2172 * OpenFlow error code as encoded by ofp_mkerr() on failure.
2174 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2177 add_flow(struct ofproto *ofproto, struct ofconn *ofconn, struct flow_mod *fm,
2178 const struct ofp_header *request)
2180 struct classifier *table;
2181 struct ofopgroup *group;
2182 struct rule *victim;
2186 /* Check for overlap, if requested. */
2187 if (fm->flags & OFPFF_CHECK_OVERLAP) {
2188 struct classifier *cls;
2190 FOR_EACH_MATCHING_TABLE (cls, fm->table_id, ofproto) {
2191 if (classifier_rule_overlaps(cls, &fm->cr)) {
2192 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2198 if (fm->table_id == 0xff) {
2200 if (ofproto->ofproto_class->rule_choose_table) {
2201 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2206 assert(table_id < ofproto->n_tables);
2207 table = &ofproto->tables[table_id];
2209 table = &ofproto->tables[0];
2211 } else if (fm->table_id < ofproto->n_tables) {
2212 table = &ofproto->tables[fm->table_id];
2214 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2217 /* Serialize against pending deletion. */
2218 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2219 return OFPROTO_POSTPONE;
2222 /* Allocate new rule. */
2223 rule = ofproto->ofproto_class->rule_alloc();
2225 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2226 ofproto->name, strerror(error));
2229 rule->ofproto = ofproto;
2231 rule->pending = NULL;
2232 rule->flow_cookie = fm->cookie;
2233 rule->created = time_msec();
2234 rule->idle_timeout = fm->idle_timeout;
2235 rule->hard_timeout = fm->hard_timeout;
2236 rule->table_id = table - ofproto->tables;
2237 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2238 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2239 rule->n_actions = fm->n_actions;
2241 /* Insert new rule. */
2242 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2243 if (victim && victim->pending) {
2244 error = OFPROTO_POSTPONE;
2247 ? ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id)
2248 : ofopgroup_create(ofproto));
2249 ofoperation_create(group, rule, OFOPERATION_ADD);
2250 rule->pending->victim = victim;
2252 error = ofproto->ofproto_class->rule_construct(rule);
2254 ofoperation_destroy(rule->pending);
2256 ofopgroup_submit(group);
2259 /* Back out if an error occurred. */
2262 classifier_replace(table, &victim->cr);
2264 classifier_remove(table, &rule->cr);
2266 ofproto_rule_destroy__(rule);
2271 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2273 /* Modifies the rules listed in 'rules', changing their actions to match those
2276 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2279 * Returns 0 on success, otherwise an OpenFlow error code. */
2281 modify_flows__(struct ofconn *ofconn, const struct flow_mod *fm,
2282 const struct ofp_header *request, struct list *rules)
2284 struct ofopgroup *group;
2287 group = ofopgroup_create_for_ofconn(ofconn, request, fm->buffer_id);
2288 LIST_FOR_EACH (rule, ofproto_node, rules) {
2289 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2290 rule->actions, rule->n_actions)) {
2291 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2292 rule->pending->actions = rule->actions;
2293 rule->pending->n_actions = rule->n_actions;
2294 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2295 rule->n_actions = fm->n_actions;
2296 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2298 rule->flow_cookie = fm->cookie;
2300 ofopgroup_submit(group);
2305 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2306 * encoded by ofp_mkerr() on failure.
2308 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2311 modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm,
2312 const struct ofp_header *request)
2314 struct ofproto *p = ofconn_get_ofproto(ofconn);
2318 error = collect_rules_loose(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2319 return (error ? error
2320 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2321 : modify_flows__(ofconn, fm, request, &rules));
2324 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2325 * code as encoded by ofp_mkerr() on failure.
2327 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2330 modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2331 const struct ofp_header *request)
2333 struct ofproto *p = ofconn_get_ofproto(ofconn);
2337 error = collect_rules_strict(p, fm->table_id, &fm->cr, OFPP_NONE, &rules);
2338 return (error ? error
2339 : list_is_empty(&rules) ? add_flow(p, ofconn, fm, request)
2340 : list_is_singleton(&rules) ? modify_flows__(ofconn, fm, request,
2345 /* OFPFC_DELETE implementation. */
2347 /* Deletes the rules listed in 'rules'.
2349 * Returns 0 on success, otherwise an OpenFlow error code. */
2351 delete_flows__(struct ofconn *ofconn, const struct ofp_header *request,
2354 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2355 struct rule *rule, *next;
2356 struct ofopgroup *group;
2358 group = ofopgroup_create_for_ofconn(ofconn, request, UINT32_MAX);
2359 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2360 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2362 ofoperation_create(group, rule, OFOPERATION_DELETE);
2363 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2364 rule->ofproto->ofproto_class->rule_destruct(rule);
2366 ofopgroup_submit(group);
2371 /* Implements OFPFC_DELETE. */
2373 delete_flows_loose(struct ofconn *ofconn, const struct flow_mod *fm,
2374 const struct ofp_header *request)
2376 struct ofproto *p = ofconn_get_ofproto(ofconn);
2380 error = collect_rules_loose(p, fm->table_id, &fm->cr, fm->out_port,
2382 return (error ? error
2383 : !list_is_empty(&rules) ? delete_flows__(ofconn, request, &rules)
2387 /* Implements OFPFC_DELETE_STRICT. */
2389 delete_flow_strict(struct ofconn *ofconn, struct flow_mod *fm,
2390 const struct ofp_header *request)
2392 struct ofproto *p = ofconn_get_ofproto(ofconn);
2396 error = collect_rules_strict(p, fm->table_id, &fm->cr, fm->out_port,
2398 return (error ? error
2399 : list_is_singleton(&rules) ? delete_flows__(ofconn, request,
2405 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2407 struct ofputil_flow_removed fr;
2409 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2414 fr.cookie = rule->flow_cookie;
2416 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2417 fr.idle_timeout = rule->idle_timeout;
2418 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2421 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2424 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2425 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2428 * ofproto implementation ->run() functions should use this function to expire
2429 * OpenFlow flows. */
2431 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2433 struct ofproto *ofproto = rule->ofproto;
2434 struct ofopgroup *group;
2436 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2438 ofproto_rule_send_removed(rule, reason);
2440 group = ofopgroup_create(ofproto);
2441 ofoperation_create(group, rule, OFOPERATION_DELETE);
2442 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2443 rule->ofproto->ofproto_class->rule_destruct(rule);
2444 ofopgroup_submit(group);
2448 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2450 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2454 error = reject_slave_controller(ofconn, "flow_mod");
2459 if (list_size(&ofproto->pending) >= 50) {
2460 return OFPROTO_POSTPONE;
2463 error = ofputil_decode_flow_mod(&fm, oh,
2464 ofconn_get_flow_mod_table_id(ofconn));
2469 /* We do not support the emergency flow cache. It will hopefully get
2470 * dropped from OpenFlow in the near future. */
2471 if (fm.flags & OFPFF_EMERG) {
2472 /* There isn't a good fit for an error code, so just state that the
2473 * flow table is full. */
2474 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2477 switch (fm.command) {
2479 return add_flow(ofproto, ofconn, &fm, oh);
2482 return modify_flows_loose(ofconn, &fm, oh);
2484 case OFPFC_MODIFY_STRICT:
2485 return modify_flow_strict(ofconn, &fm, oh);
2488 return delete_flows_loose(ofconn, &fm, oh);
2490 case OFPFC_DELETE_STRICT:
2491 return delete_flow_strict(ofconn, &fm, oh);
2494 if (fm.command > 0xff) {
2495 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2496 "flow_mod_table_id extension is not enabled");
2498 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2503 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2505 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2506 struct nx_role_request *reply;
2510 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2511 VLOG_WARN_RL(&rl, "ignoring role request on service connection");
2512 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2515 role = ntohl(nrr->role);
2516 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2517 && role != NX_ROLE_SLAVE) {
2518 VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role);
2520 /* There's no good error code for this. */
2521 return ofp_mkerr(OFPET_BAD_REQUEST, -1);
2524 if (ofconn_get_role(ofconn) != role
2525 && ofconn_has_pending_opgroups(ofconn)) {
2526 return OFPROTO_POSTPONE;
2529 ofconn_set_role(ofconn, role);
2531 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2532 reply->role = htonl(role);
2533 ofconn_send_reply(ofconn, buf);
2539 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2540 const struct ofp_header *oh)
2542 const struct nxt_flow_mod_table_id *msg
2543 = (const struct nxt_flow_mod_table_id *) oh;
2545 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2550 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2552 const struct nxt_set_flow_format *msg
2553 = (const struct nxt_set_flow_format *) oh;
2556 format = ntohl(msg->format);
2557 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2558 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2561 if (format != ofconn_get_flow_format(ofconn)
2562 && ofconn_has_pending_opgroups(ofconn)) {
2563 /* Avoid sending async messages in surprising flow format. */
2564 return OFPROTO_POSTPONE;
2567 ofconn_set_flow_format(ofconn, format);
2572 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2574 struct ofp_header *ob;
2577 if (ofconn_has_pending_opgroups(ofconn)) {
2578 return OFPROTO_POSTPONE;
2581 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2582 ofconn_send_reply(ofconn, buf);
2587 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2589 const struct ofp_header *oh = msg->data;
2590 const struct ofputil_msg_type *type;
2593 error = ofputil_decode_msg_type(oh, &type);
2598 switch (ofputil_msg_type_code(type)) {
2599 /* OpenFlow requests. */
2600 case OFPUTIL_OFPT_ECHO_REQUEST:
2601 return handle_echo_request(ofconn, oh);
2603 case OFPUTIL_OFPT_FEATURES_REQUEST:
2604 return handle_features_request(ofconn, oh);
2606 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2607 return handle_get_config_request(ofconn, oh);
2609 case OFPUTIL_OFPT_SET_CONFIG:
2610 return handle_set_config(ofconn, msg->data);
2612 case OFPUTIL_OFPT_PACKET_OUT:
2613 return handle_packet_out(ofconn, oh);
2615 case OFPUTIL_OFPT_PORT_MOD:
2616 return handle_port_mod(ofconn, oh);
2618 case OFPUTIL_OFPT_FLOW_MOD:
2619 return handle_flow_mod(ofconn, oh);
2621 case OFPUTIL_OFPT_BARRIER_REQUEST:
2622 return handle_barrier_request(ofconn, oh);
2624 /* OpenFlow replies. */
2625 case OFPUTIL_OFPT_ECHO_REPLY:
2628 /* Nicira extension requests. */
2629 case OFPUTIL_NXT_ROLE_REQUEST:
2630 return handle_role_request(ofconn, oh);
2632 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2633 return handle_nxt_flow_mod_table_id(ofconn, oh);
2635 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2636 return handle_nxt_set_flow_format(ofconn, oh);
2638 case OFPUTIL_NXT_FLOW_MOD:
2639 return handle_flow_mod(ofconn, oh);
2641 /* Statistics requests. */
2642 case OFPUTIL_OFPST_DESC_REQUEST:
2643 return handle_desc_stats_request(ofconn, msg->data);
2645 case OFPUTIL_OFPST_FLOW_REQUEST:
2646 case OFPUTIL_NXST_FLOW_REQUEST:
2647 return handle_flow_stats_request(ofconn, msg->data);
2649 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2650 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2651 return handle_aggregate_stats_request(ofconn, msg->data);
2653 case OFPUTIL_OFPST_TABLE_REQUEST:
2654 return handle_table_stats_request(ofconn, msg->data);
2656 case OFPUTIL_OFPST_PORT_REQUEST:
2657 return handle_port_stats_request(ofconn, msg->data);
2659 case OFPUTIL_OFPST_QUEUE_REQUEST:
2660 return handle_queue_stats_request(ofconn, msg->data);
2662 case OFPUTIL_MSG_INVALID:
2663 case OFPUTIL_OFPT_HELLO:
2664 case OFPUTIL_OFPT_ERROR:
2665 case OFPUTIL_OFPT_FEATURES_REPLY:
2666 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2667 case OFPUTIL_OFPT_PACKET_IN:
2668 case OFPUTIL_OFPT_FLOW_REMOVED:
2669 case OFPUTIL_OFPT_PORT_STATUS:
2670 case OFPUTIL_OFPT_BARRIER_REPLY:
2671 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2672 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2673 case OFPUTIL_OFPST_DESC_REPLY:
2674 case OFPUTIL_OFPST_FLOW_REPLY:
2675 case OFPUTIL_OFPST_QUEUE_REPLY:
2676 case OFPUTIL_OFPST_PORT_REPLY:
2677 case OFPUTIL_OFPST_TABLE_REPLY:
2678 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2679 case OFPUTIL_NXT_ROLE_REPLY:
2680 case OFPUTIL_NXT_FLOW_REMOVED:
2681 case OFPUTIL_NXST_FLOW_REPLY:
2682 case OFPUTIL_NXST_AGGREGATE_REPLY:
2684 if (VLOG_IS_WARN_ENABLED()) {
2685 char *s = ofp_to_string(oh, ntohs(oh->length), 2);
2686 VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s);
2689 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2690 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2692 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2698 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2700 int error = handle_openflow__(ofconn, ofp_msg);
2701 if (error && error != OFPROTO_POSTPONE) {
2702 ofconn_send_error(ofconn, ofp_msg->data, error);
2704 COVERAGE_INC(ofproto_recv_openflow);
2705 return error != OFPROTO_POSTPONE;
2708 /* Asynchronous operations. */
2710 /* Creates and returns a new ofopgroup that is not associated with any
2711 * OpenFlow connection.
2713 * The caller should add operations to the returned group with
2714 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2715 static struct ofopgroup *
2716 ofopgroup_create(struct ofproto *ofproto)
2718 struct ofopgroup *group = xzalloc(sizeof *group);
2719 group->ofproto = ofproto;
2720 list_init(&group->ofproto_node);
2721 list_init(&group->ops);
2722 list_init(&group->ofconn_node);
2726 /* Creates and returns a new ofopgroup that is associated with 'ofconn'. If
2727 * the ofopgroup eventually fails, then the error reply will include 'request'.
2728 * If the ofopgroup eventually succeeds, then the packet with buffer id
2729 * 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
2731 * The caller should add operations to the returned group with
2732 * ofoperation_create() and then submit it with ofopgroup_submit(). */
2733 static struct ofopgroup *
2734 ofopgroup_create_for_ofconn(struct ofconn *ofconn,
2735 const struct ofp_header *request,
2738 struct ofopgroup *group = ofopgroup_create(ofconn_get_ofproto(ofconn));
2739 size_t request_len = ntohs(request->length);
2741 ofconn_add_opgroup(ofconn, &group->ofconn_node);
2742 group->ofconn = ofconn;
2743 group->request = xmemdup(request, MIN(request_len, 64));
2744 group->buffer_id = buffer_id;
2749 /* Submits 'group' for processing.
2751 * If 'group' contains no operations (e.g. none were ever added, or all of the
2752 * ones that were added completed synchronously), then it is destroyed
2753 * immediately. Otherwise it is added to the ofproto's list of pending
2756 ofopgroup_submit(struct ofopgroup *group)
2758 if (list_is_empty(&group->ops)) {
2759 ofopgroup_destroy(group);
2761 list_push_back(&group->ofproto->pending, &group->ofproto_node);
2766 ofopgroup_destroy(struct ofopgroup *group)
2768 assert(list_is_empty(&group->ops));
2769 if (!list_is_empty(&group->ofproto_node)) {
2770 list_remove(&group->ofproto_node);
2772 if (!list_is_empty(&group->ofconn_node)) {
2773 list_remove(&group->ofconn_node);
2775 ofconn_send_error(group->ofconn, group->request, group->error);
2777 connmgr_retry(group->ofproto->connmgr);
2779 free(group->request);
2783 /* Initiates a new operation on 'rule', of the specified 'type', within
2784 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
2786 ofoperation_create(struct ofopgroup *group, struct rule *rule,
2787 enum ofoperation_type type)
2789 struct ofoperation *op;
2791 assert(!rule->pending);
2793 op = rule->pending = xzalloc(sizeof *op);
2795 list_push_back(&group->ops, &op->group_node);
2799 op->flow_cookie = rule->flow_cookie;
2801 if (type == OFOPERATION_DELETE) {
2802 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
2803 cls_rule_hash(&rule->cr, rule->table_id));
2808 ofoperation_destroy(struct ofoperation *op)
2810 struct ofopgroup *group = op->group;
2813 op->rule->pending = NULL;
2815 if (op->type == OFOPERATION_DELETE) {
2816 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
2818 list_remove(&op->group_node);
2822 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
2823 ofopgroup_destroy(group);
2827 /* Indicates that 'op' completed with status 'error', which is either 0 to
2828 * indicate success or an OpenFlow error code (constructed with
2829 * e.g. ofp_mkerr()).
2831 * If 'op' is a "delete flow" operation, 'error' must be 0. That is, flow
2832 * deletions are not allowed to fail.
2834 * Please see the large comment in ofproto/ofproto-provider.h titled
2835 * "Asynchronous Operation Support" for more information. */
2837 ofoperation_complete(struct ofoperation *op, int error)
2839 struct ofopgroup *group = op->group;
2840 struct rule *rule = op->rule;
2841 struct classifier *table = &rule->ofproto->tables[rule->table_id];
2843 assert(rule->pending == op);
2844 assert(op->status < 0);
2849 && op->type != OFOPERATION_DELETE
2851 && group->buffer_id != UINT32_MAX
2852 && list_is_singleton(&op->group_node)) {
2853 struct ofpbuf *packet;
2856 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
2860 error = rule_execute(rule, in_port, packet);
2863 if (!group->error) {
2864 group->error = error;
2868 case OFOPERATION_ADD:
2871 ofproto_rule_destroy__(op->victim);
2875 classifier_replace(table, &op->victim->cr);
2878 classifier_remove(table, &rule->cr);
2880 ofproto_rule_destroy__(rule);
2885 case OFOPERATION_DELETE:
2887 ofproto_rule_destroy__(rule);
2891 case OFOPERATION_MODIFY:
2893 free(rule->actions);
2894 rule->actions = op->actions;
2895 rule->n_actions = op->n_actions;
2903 ofoperation_destroy(op);
2907 ofoperation_get_victim(struct ofoperation *op)
2909 assert(op->type == OFOPERATION_ADD);
2914 pick_datapath_id(const struct ofproto *ofproto)
2916 const struct ofport *port;
2918 port = ofproto_get_port(ofproto, OFPP_LOCAL);
2920 uint8_t ea[ETH_ADDR_LEN];
2923 error = netdev_get_etheraddr(port->netdev, ea);
2925 return eth_addr_to_uint64(ea);
2927 VLOG_WARN("could not get MAC address for %s (%s)",
2928 netdev_get_name(port->netdev), strerror(error));
2930 return ofproto->fallback_dpid;
2934 pick_fallback_dpid(void)
2936 uint8_t ea[ETH_ADDR_LEN];
2937 eth_addr_nicira_random(ea);
2938 return eth_addr_to_uint64(ea);
2941 /* unixctl commands. */
2944 ofproto_lookup(const char *name)
2946 struct ofproto *ofproto;
2948 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
2950 if (!strcmp(ofproto->name, name)) {
2958 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
2959 void *aux OVS_UNUSED)
2961 struct ofproto *ofproto;
2965 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
2966 ds_put_format(&results, "%s\n", ofproto->name);
2968 unixctl_command_reply(conn, 200, ds_cstr(&results));
2969 ds_destroy(&results);
2973 ofproto_unixctl_init(void)
2975 static bool registered;
2981 unixctl_command_register("ofproto/list", ofproto_unixctl_list, NULL);