2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc.
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.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
32 #include "meta-flow.h"
35 #include "ofp-actions.h"
36 #include "ofp-errors.h"
38 #include "ofp-print.h"
41 #include "ofproto-provider.h"
42 #include "openflow/nicira-ext.h"
43 #include "openflow/openflow.h"
47 #include "poll-loop.h"
53 #include "unaligned.h"
57 VLOG_DEFINE_THIS_MODULE(ofproto);
59 COVERAGE_DEFINE(ofproto_error);
60 COVERAGE_DEFINE(ofproto_flush);
61 COVERAGE_DEFINE(ofproto_no_packet_in);
62 COVERAGE_DEFINE(ofproto_packet_out);
63 COVERAGE_DEFINE(ofproto_queue_req);
64 COVERAGE_DEFINE(ofproto_recv_openflow);
65 COVERAGE_DEFINE(ofproto_reinit_ports);
66 COVERAGE_DEFINE(ofproto_uninstallable);
67 COVERAGE_DEFINE(ofproto_update_port);
70 S_OPENFLOW, /* Processing OpenFlow commands. */
71 S_EVICT, /* Evicting flows from over-limit tables. */
72 S_FLUSH, /* Deleting all flow table rules. */
75 enum ofoperation_type {
81 /* A single OpenFlow request can execute any number of operations. The
82 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
83 * ofconn to which an error reply should be sent if necessary.
85 * ofproto initiates some operations internally. These operations are still
86 * assigned to groups but will not have an associated ofconn. */
88 struct ofproto *ofproto; /* Owning ofproto. */
89 struct list ofproto_node; /* In ofproto's "pending" list. */
90 struct list ops; /* List of "struct ofoperation"s. */
91 int n_running; /* Number of ops still pending. */
93 /* Data needed to send OpenFlow reply on failure or to send a buffered
96 * If list_is_empty(ofconn_node) then this ofopgroup never had an
97 * associated ofconn or its ofconn's connection dropped after it initiated
98 * the operation. In the latter case 'ofconn' is a wild pointer that
99 * refers to freed memory, so the 'ofconn' member must be used only if
100 * !list_is_empty(ofconn_node).
102 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
103 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
104 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
105 uint32_t buffer_id; /* Buffer id from original request. */
108 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
109 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
110 const struct ofp_header *,
112 static void ofopgroup_submit(struct ofopgroup *);
113 static void ofopgroup_complete(struct ofopgroup *);
115 /* A single flow table operation. */
117 struct ofopgroup *group; /* Owning group. */
118 struct list group_node; /* In ofopgroup's "ops" list. */
119 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
120 struct rule *rule; /* Rule being operated upon. */
121 enum ofoperation_type type; /* Type of operation. */
123 /* OFOPERATION_ADD. */
124 struct rule *victim; /* Rule being replaced, if any.. */
126 /* OFOPERATION_MODIFY: The old actions, if the actions are changing. */
127 struct ofpact *ofpacts;
130 /* OFOPERATION_DELETE. */
131 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
133 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
134 enum ofperr error; /* 0 if no error. */
137 static struct ofoperation *ofoperation_create(struct ofopgroup *,
139 enum ofoperation_type,
140 enum ofp_flow_removed_reason);
141 static void ofoperation_destroy(struct ofoperation *);
144 static void oftable_init(struct oftable *);
145 static void oftable_destroy(struct oftable *);
147 static void oftable_set_name(struct oftable *, const char *name);
149 static void oftable_disable_eviction(struct oftable *);
150 static void oftable_enable_eviction(struct oftable *,
151 const struct mf_subfield *fields,
154 static void oftable_remove_rule(struct rule *);
155 static struct rule *oftable_replace_rule(struct rule *);
156 static void oftable_substitute_rule(struct rule *old, struct rule *new);
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
163 * An oftable owns any number of eviction groups, each of which contains any
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
180 static struct rule *choose_rule_to_evict(struct oftable *);
181 static void ofproto_evict(struct ofproto *);
182 static uint32_t rule_eviction_priority(struct rule *);
185 static void ofport_destroy__(struct ofport *);
186 static void ofport_destroy(struct ofport *);
188 static void update_port(struct ofproto *, const char *devname);
189 static int init_ports(struct ofproto *);
190 static void reinit_ports(struct ofproto *);
193 static void ofproto_rule_destroy__(struct rule *);
194 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
195 static bool rule_is_modifiable(const struct rule *);
198 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
199 const struct ofputil_flow_mod *,
200 const struct ofp_header *);
201 static void delete_flow__(struct rule *, struct ofopgroup *);
202 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
203 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
204 const struct ofputil_flow_mod *,
205 const struct ofp_header *);
208 static uint64_t pick_datapath_id(const struct ofproto *);
209 static uint64_t pick_fallback_dpid(void);
210 static void ofproto_destroy__(struct ofproto *);
211 static void update_mtu(struct ofproto *, struct ofport *);
214 static void ofproto_unixctl_init(void);
216 /* All registered ofproto classes, in probe order. */
217 static const struct ofproto_class **ofproto_classes;
218 static size_t n_ofproto_classes;
219 static size_t allocated_ofproto_classes;
221 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
222 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
227 ofproto_initialize(void)
233 ofproto_class_register(&ofproto_dpif_class);
237 /* 'type' should be a normalized datapath type, as returned by
238 * ofproto_normalize_type(). Returns the corresponding ofproto_class
239 * structure, or a null pointer if there is none registered for 'type'. */
240 static const struct ofproto_class *
241 ofproto_class_find__(const char *type)
245 ofproto_initialize();
246 for (i = 0; i < n_ofproto_classes; i++) {
247 const struct ofproto_class *class = ofproto_classes[i];
252 class->enumerate_types(&types);
253 found = sset_contains(&types, type);
254 sset_destroy(&types);
260 VLOG_WARN("unknown datapath type %s", type);
264 /* Registers a new ofproto class. After successful registration, new ofprotos
265 * of that type can be created using ofproto_create(). */
267 ofproto_class_register(const struct ofproto_class *new_class)
271 for (i = 0; i < n_ofproto_classes; i++) {
272 if (ofproto_classes[i] == new_class) {
277 if (n_ofproto_classes >= allocated_ofproto_classes) {
278 ofproto_classes = x2nrealloc(ofproto_classes,
279 &allocated_ofproto_classes,
280 sizeof *ofproto_classes);
282 ofproto_classes[n_ofproto_classes++] = new_class;
286 /* Unregisters a datapath provider. 'type' must have been previously
287 * registered and not currently be in use by any ofprotos. After
288 * unregistration new datapaths of that type cannot be opened using
289 * ofproto_create(). */
291 ofproto_class_unregister(const struct ofproto_class *class)
295 for (i = 0; i < n_ofproto_classes; i++) {
296 if (ofproto_classes[i] == class) {
297 for (i++; i < n_ofproto_classes; i++) {
298 ofproto_classes[i - 1] = ofproto_classes[i];
304 VLOG_WARN("attempted to unregister an ofproto class that is not "
309 /* Clears 'types' and enumerates all registered ofproto types into it. The
310 * caller must first initialize the sset. */
312 ofproto_enumerate_types(struct sset *types)
316 ofproto_initialize();
317 for (i = 0; i < n_ofproto_classes; i++) {
318 ofproto_classes[i]->enumerate_types(types);
322 /* Returns the fully spelled out name for the given ofproto 'type'.
324 * Normalized type string can be compared with strcmp(). Unnormalized type
325 * string might be the same even if they have different spellings. */
327 ofproto_normalize_type(const char *type)
329 return type && type[0] ? type : "system";
332 /* Clears 'names' and enumerates the names of all known created ofprotos with
333 * the given 'type'. The caller must first initialize the sset. Returns 0 if
334 * successful, otherwise a positive errno value.
336 * Some kinds of datapaths might not be practically enumerable. This is not
337 * considered an error. */
339 ofproto_enumerate_names(const char *type, struct sset *names)
341 const struct ofproto_class *class = ofproto_class_find__(type);
342 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
346 ofproto_create(const char *datapath_name, const char *datapath_type,
347 struct ofproto **ofprotop)
349 const struct ofproto_class *class;
350 struct ofproto *ofproto;
355 ofproto_initialize();
356 ofproto_unixctl_init();
358 datapath_type = ofproto_normalize_type(datapath_type);
359 class = ofproto_class_find__(datapath_type);
361 VLOG_WARN("could not create datapath %s of unknown type %s",
362 datapath_name, datapath_type);
366 ofproto = class->alloc();
368 VLOG_ERR("failed to allocate datapath %s of type %s",
369 datapath_name, datapath_type);
374 memset(ofproto, 0, sizeof *ofproto);
375 ofproto->ofproto_class = class;
376 ofproto->name = xstrdup(datapath_name);
377 ofproto->type = xstrdup(datapath_type);
378 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
379 hash_string(ofproto->name, 0));
380 ofproto->datapath_id = 0;
381 ofproto_set_flow_eviction_threshold(ofproto,
382 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
383 ofproto->forward_bpdu = false;
384 ofproto->fallback_dpid = pick_fallback_dpid();
385 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
386 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
387 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
388 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
389 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
390 ofproto->frag_handling = OFPC_FRAG_NORMAL;
391 hmap_init(&ofproto->ports);
392 shash_init(&ofproto->port_by_name);
393 ofproto->tables = NULL;
394 ofproto->n_tables = 0;
395 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
396 ofproto->state = S_OPENFLOW;
397 list_init(&ofproto->pending);
398 ofproto->n_pending = 0;
399 hmap_init(&ofproto->deletions);
400 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
401 ofproto->first_op = ofproto->last_op = LLONG_MIN;
402 ofproto->next_op_report = LLONG_MAX;
403 ofproto->op_backoff = LLONG_MIN;
404 ofproto->vlan_bitmap = NULL;
405 ofproto->vlans_changed = false;
406 ofproto->min_mtu = INT_MAX;
408 error = ofproto->ofproto_class->construct(ofproto);
410 VLOG_ERR("failed to open datapath %s: %s",
411 datapath_name, strerror(error));
412 ofproto_destroy__(ofproto);
416 assert(ofproto->n_tables);
418 ofproto->datapath_id = pick_datapath_id(ofproto);
426 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
428 struct oftable *table;
430 assert(!ofproto->n_tables);
431 assert(n_tables >= 1 && n_tables <= 255);
433 ofproto->n_tables = n_tables;
434 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
435 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
441 ofproto_get_datapath_id(const struct ofproto *ofproto)
443 return ofproto->datapath_id;
447 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
449 uint64_t old_dpid = p->datapath_id;
450 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
451 if (p->datapath_id != old_dpid) {
452 /* Force all active connections to reconnect, since there is no way to
453 * notify a controller that the datapath ID has changed. */
454 ofproto_reconnect_controllers(p);
459 ofproto_set_controllers(struct ofproto *p,
460 const struct ofproto_controller *controllers,
461 size_t n_controllers)
463 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
467 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
469 connmgr_set_fail_mode(p->connmgr, fail_mode);
472 /* Drops the connections between 'ofproto' and all of its controllers, forcing
473 * them to reconnect. */
475 ofproto_reconnect_controllers(struct ofproto *ofproto)
477 connmgr_reconnect(ofproto->connmgr);
480 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
481 * in-band control should guarantee access, in the same way that in-band
482 * control guarantees access to OpenFlow controllers. */
484 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
485 const struct sockaddr_in *extras, size_t n)
487 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
490 /* Sets the OpenFlow queue used by flows set up by in-band control on
491 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
492 * flows will use the default queue. */
494 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
496 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
499 /* Sets the number of flows at which eviction from the kernel flow table
502 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
504 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
505 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
507 ofproto->flow_eviction_threshold = threshold;
511 /* If forward_bpdu is true, the NORMAL action will forward frames with
512 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
513 * the NORMAL action will drop these frames. */
515 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
517 bool old_val = ofproto->forward_bpdu;
518 ofproto->forward_bpdu = forward_bpdu;
519 if (old_val != ofproto->forward_bpdu) {
520 if (ofproto->ofproto_class->forward_bpdu_changed) {
521 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
526 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
527 * 'idle_time', in seconds. */
529 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
531 if (ofproto->ofproto_class->set_mac_idle_time) {
532 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
537 ofproto_set_desc(struct ofproto *p,
538 const char *mfr_desc, const char *hw_desc,
539 const char *sw_desc, const char *serial_desc,
542 struct ofp_desc_stats *ods;
545 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
546 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
547 p->name, sizeof ods->mfr_desc);
550 p->mfr_desc = xstrdup(mfr_desc);
553 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
554 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
555 p->name, sizeof ods->hw_desc);
558 p->hw_desc = xstrdup(hw_desc);
561 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
562 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
563 p->name, sizeof ods->sw_desc);
566 p->sw_desc = xstrdup(sw_desc);
569 if (strlen(serial_desc) >= sizeof ods->serial_num) {
570 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
571 "bytes", p->name, sizeof ods->serial_num);
573 free(p->serial_desc);
574 p->serial_desc = xstrdup(serial_desc);
577 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
578 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
579 p->name, sizeof ods->dp_desc);
582 p->dp_desc = xstrdup(dp_desc);
587 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
589 return connmgr_set_snoops(ofproto->connmgr, snoops);
593 ofproto_set_netflow(struct ofproto *ofproto,
594 const struct netflow_options *nf_options)
596 if (nf_options && sset_is_empty(&nf_options->collectors)) {
600 if (ofproto->ofproto_class->set_netflow) {
601 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
603 return nf_options ? EOPNOTSUPP : 0;
608 ofproto_set_sflow(struct ofproto *ofproto,
609 const struct ofproto_sflow_options *oso)
611 if (oso && sset_is_empty(&oso->targets)) {
615 if (ofproto->ofproto_class->set_sflow) {
616 return ofproto->ofproto_class->set_sflow(ofproto, oso);
618 return oso ? EOPNOTSUPP : 0;
622 /* Spanning Tree Protocol (STP) configuration. */
624 /* Configures STP on 'ofproto' using the settings defined in 's'. If
625 * 's' is NULL, disables STP.
627 * Returns 0 if successful, otherwise a positive errno value. */
629 ofproto_set_stp(struct ofproto *ofproto,
630 const struct ofproto_stp_settings *s)
632 return (ofproto->ofproto_class->set_stp
633 ? ofproto->ofproto_class->set_stp(ofproto, s)
637 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
638 * 'enabled' member of 's' is false, then the other members are not
641 * Returns 0 if successful, otherwise a positive errno value. */
643 ofproto_get_stp_status(struct ofproto *ofproto,
644 struct ofproto_stp_status *s)
646 return (ofproto->ofproto_class->get_stp_status
647 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
651 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
652 * in 's'. The caller is responsible for assigning STP port numbers
653 * (using the 'port_num' member in the range of 1 through 255, inclusive)
654 * and ensuring there are no duplicates. If the 's' is NULL, then STP
655 * is disabled on the port.
657 * Returns 0 if successful, otherwise a positive errno value.*/
659 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
660 const struct ofproto_port_stp_settings *s)
662 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
664 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
665 ofproto->name, ofp_port);
669 return (ofproto->ofproto_class->set_stp_port
670 ? ofproto->ofproto_class->set_stp_port(ofport, s)
674 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
675 * 's'. If the 'enabled' member in 's' is false, then the other members
676 * are not meaningful.
678 * Returns 0 if successful, otherwise a positive errno value.*/
680 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
681 struct ofproto_port_stp_status *s)
683 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
685 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
686 "port %"PRIu16, ofproto->name, ofp_port);
690 return (ofproto->ofproto_class->get_stp_port_status
691 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
695 /* Queue DSCP configuration. */
697 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
698 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
699 * to implement QoS. Instead, it is used to implement features which require
700 * knowledge of what queues exist on a port, and some basic information about
703 * Returns 0 if successful, otherwise a positive errno value. */
705 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
706 const struct ofproto_port_queue *queues,
709 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
712 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
713 ofproto->name, ofp_port);
717 return (ofproto->ofproto_class->set_queues
718 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
722 /* Connectivity Fault Management configuration. */
724 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
726 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
728 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
729 if (ofport && ofproto->ofproto_class->set_cfm) {
730 ofproto->ofproto_class->set_cfm(ofport, NULL);
734 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
735 * basic configuration from the configuration members in 'cfm', and the remote
736 * maintenance point ID from remote_mpid. Ignores the statistics members of
739 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
741 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
742 const struct cfm_settings *s)
744 struct ofport *ofport;
747 ofport = ofproto_get_port(ofproto, ofp_port);
749 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
750 ofproto->name, ofp_port);
754 /* XXX: For configuration simplicity, we only support one remote_mpid
755 * outside of the CFM module. It's not clear if this is the correct long
756 * term solution or not. */
757 error = (ofproto->ofproto_class->set_cfm
758 ? ofproto->ofproto_class->set_cfm(ofport, s)
761 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
762 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
767 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
768 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
769 * 0 if LACP partner information is not current (generally indicating a
770 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
772 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
774 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
775 return (ofport && ofproto->ofproto_class->port_is_lacp_current
776 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
782 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
783 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
784 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
785 * configuration plus, if there is more than one slave, a bonding
788 * If 'aux' is already registered then this function updates its configuration
789 * to 's'. Otherwise, this function registers a new bundle.
791 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
794 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
795 const struct ofproto_bundle_settings *s)
797 return (ofproto->ofproto_class->bundle_set
798 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
802 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
803 * If no such bundle has been registered, this has no effect. */
805 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
807 return ofproto_bundle_register(ofproto, aux, NULL);
811 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
812 * If 'aux' is already registered then this function updates its configuration
813 * to 's'. Otherwise, this function registers a new mirror. */
815 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
816 const struct ofproto_mirror_settings *s)
818 return (ofproto->ofproto_class->mirror_set
819 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
823 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
824 * If no mirror has been registered, this has no effect. */
826 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
828 return ofproto_mirror_register(ofproto, aux, NULL);
831 /* Retrieves statistics from mirror associated with client data pointer
832 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
833 * 'bytes', respectively. If a particular counters is not supported,
834 * the appropriate argument is set to UINT64_MAX. */
836 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
837 uint64_t *packets, uint64_t *bytes)
839 if (!ofproto->ofproto_class->mirror_get_stats) {
840 *packets = *bytes = UINT64_MAX;
844 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
848 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
849 * which all packets are flooded, instead of using MAC learning. If
850 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
852 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
855 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
857 return (ofproto->ofproto_class->set_flood_vlans
858 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
862 /* Returns true if 'aux' is a registered bundle that is currently in use as the
863 * output for a mirror. */
865 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
867 return (ofproto->ofproto_class->is_mirror_output_bundle
868 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
872 /* Configuration of OpenFlow tables. */
874 /* Returns the number of OpenFlow tables in 'ofproto'. */
876 ofproto_get_n_tables(const struct ofproto *ofproto)
878 return ofproto->n_tables;
881 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
882 * settings from 's'. 'table_id' must be in the range 0 through the number of
883 * OpenFlow tables in 'ofproto' minus 1, inclusive.
885 * For read-only tables, only the name may be configured. */
887 ofproto_configure_table(struct ofproto *ofproto, int table_id,
888 const struct ofproto_table_settings *s)
890 struct oftable *table;
892 assert(table_id >= 0 && table_id < ofproto->n_tables);
893 table = &ofproto->tables[table_id];
895 oftable_set_name(table, s->name);
897 if (table->flags & OFTABLE_READONLY) {
902 oftable_enable_eviction(table, s->groups, s->n_groups);
904 oftable_disable_eviction(table);
907 table->max_flows = s->max_flows;
908 if (classifier_count(&table->cls) > table->max_flows
909 && table->eviction_fields) {
910 /* 'table' contains more flows than allowed. We might not be able to
911 * evict them right away because of the asynchronous nature of flow
912 * table changes. Schedule eviction for later. */
913 switch (ofproto->state) {
915 ofproto->state = S_EVICT;
919 /* We're already deleting flows, nothing more to do. */
926 ofproto_has_snoops(const struct ofproto *ofproto)
928 return connmgr_has_snoops(ofproto->connmgr);
932 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
934 connmgr_get_snoops(ofproto->connmgr, snoops);
938 ofproto_flush__(struct ofproto *ofproto)
940 struct ofopgroup *group;
941 struct oftable *table;
943 if (ofproto->ofproto_class->flush) {
944 ofproto->ofproto_class->flush(ofproto);
947 group = ofopgroup_create_unattached(ofproto);
948 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
949 struct rule *rule, *next_rule;
950 struct cls_cursor cursor;
952 if (table->flags & OFTABLE_HIDDEN) {
956 cls_cursor_init(&cursor, &table->cls, NULL);
957 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
958 if (!rule->pending) {
959 ofoperation_create(group, rule, OFOPERATION_DELETE,
961 oftable_remove_rule(rule);
962 ofproto->ofproto_class->rule_destruct(rule);
966 ofopgroup_submit(group);
970 ofproto_destroy__(struct ofproto *ofproto)
972 struct oftable *table;
974 assert(list_is_empty(&ofproto->pending));
975 assert(!ofproto->n_pending);
977 connmgr_destroy(ofproto->connmgr);
979 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
982 free(ofproto->mfr_desc);
983 free(ofproto->hw_desc);
984 free(ofproto->sw_desc);
985 free(ofproto->serial_desc);
986 free(ofproto->dp_desc);
987 hmap_destroy(&ofproto->ports);
988 shash_destroy(&ofproto->port_by_name);
990 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
991 oftable_destroy(table);
993 free(ofproto->tables);
995 hmap_destroy(&ofproto->deletions);
997 free(ofproto->vlan_bitmap);
999 ofproto->ofproto_class->dealloc(ofproto);
1003 ofproto_destroy(struct ofproto *p)
1005 struct ofport *ofport, *next_ofport;
1012 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1013 ofport_destroy(ofport);
1016 p->ofproto_class->destruct(p);
1017 ofproto_destroy__(p);
1020 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1021 * kernel datapath, for example, this destroys the datapath in the kernel, and
1022 * with the netdev-based datapath, it tears down the data structures that
1023 * represent the datapath.
1025 * The datapath should not be currently open as an ofproto. */
1027 ofproto_delete(const char *name, const char *type)
1029 const struct ofproto_class *class = ofproto_class_find__(type);
1030 return (!class ? EAFNOSUPPORT
1031 : !class->del ? EACCES
1032 : class->del(type, name));
1036 process_port_change(struct ofproto *ofproto, int error, char *devname)
1038 if (error == ENOBUFS) {
1039 reinit_ports(ofproto);
1040 } else if (!error) {
1041 update_port(ofproto, devname);
1047 ofproto_run(struct ofproto *p)
1049 struct sset changed_netdevs;
1050 const char *changed_netdev;
1051 struct ofport *ofport;
1054 error = p->ofproto_class->run(p);
1055 if (error && error != EAGAIN) {
1056 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1059 if (p->ofproto_class->port_poll) {
1062 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1063 process_port_change(p, error, devname);
1067 /* Update OpenFlow port status for any port whose netdev has changed.
1069 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1070 * destroyed, so it's not safe to update ports directly from the
1071 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1072 * need this two-phase approach. */
1073 sset_init(&changed_netdevs);
1074 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1075 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1076 if (ofport->change_seq != change_seq) {
1077 ofport->change_seq = change_seq;
1078 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1081 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1082 update_port(p, changed_netdev);
1084 sset_destroy(&changed_netdevs);
1088 connmgr_run(p->connmgr, handle_openflow);
1092 connmgr_run(p->connmgr, NULL);
1094 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1095 p->state = S_OPENFLOW;
1100 connmgr_run(p->connmgr, NULL);
1102 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1103 connmgr_flushed(p->connmgr);
1104 p->state = S_OPENFLOW;
1112 if (time_msec() >= p->next_op_report) {
1113 long long int ago = (time_msec() - p->first_op) / 1000;
1114 long long int interval = (p->last_op - p->first_op) / 1000;
1118 ds_put_format(&s, "%d flow_mods ",
1119 p->n_add + p->n_delete + p->n_modify);
1120 if (interval == ago) {
1121 ds_put_format(&s, "in the last %lld s", ago);
1122 } else if (interval) {
1123 ds_put_format(&s, "in the %lld s starting %lld s ago",
1126 ds_put_format(&s, "%lld s ago", ago);
1129 ds_put_cstr(&s, " (");
1131 ds_put_format(&s, "%d adds, ", p->n_add);
1134 ds_put_format(&s, "%d deletes, ", p->n_delete);
1137 ds_put_format(&s, "%d modifications, ", p->n_modify);
1140 ds_put_char(&s, ')');
1142 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1145 p->n_add = p->n_delete = p->n_modify = 0;
1146 p->next_op_report = LLONG_MAX;
1152 /* Performs periodic activity required by 'ofproto' that needs to be done
1153 * with the least possible latency.
1155 * It makes sense to call this function a couple of times per poll loop, to
1156 * provide a significant performance boost on some benchmarks with the
1157 * ofproto-dpif implementation. */
1159 ofproto_run_fast(struct ofproto *p)
1163 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1164 if (error && error != EAGAIN) {
1165 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1166 p->name, strerror(error));
1172 ofproto_wait(struct ofproto *p)
1174 struct ofport *ofport;
1176 p->ofproto_class->wait(p);
1177 if (p->ofproto_class->port_poll_wait) {
1178 p->ofproto_class->port_poll_wait(p);
1181 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1182 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1183 poll_immediate_wake();
1189 connmgr_wait(p->connmgr, true);
1194 connmgr_wait(p->connmgr, false);
1195 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1196 poll_immediate_wake();
1203 ofproto_is_alive(const struct ofproto *p)
1205 return connmgr_has_controllers(p->connmgr);
1208 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1209 * memory_report(). */
1211 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1213 const struct oftable *table;
1214 unsigned int n_rules;
1216 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1217 simap_increase(usage, "ops",
1218 ofproto->n_pending + hmap_count(&ofproto->deletions));
1221 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1222 n_rules += classifier_count(&table->cls);
1224 simap_increase(usage, "rules", n_rules);
1226 if (ofproto->ofproto_class->get_memory_usage) {
1227 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1230 connmgr_get_memory_usage(ofproto->connmgr, usage);
1234 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1237 connmgr_get_controller_info(ofproto->connmgr, info);
1241 ofproto_free_ofproto_controller_info(struct shash *info)
1243 connmgr_free_controller_info(info);
1246 /* Makes a deep copy of 'old' into 'port'. */
1248 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1250 port->name = xstrdup(old->name);
1251 port->type = xstrdup(old->type);
1252 port->ofp_port = old->ofp_port;
1255 /* Frees memory allocated to members of 'ofproto_port'.
1257 * Do not call this function on an ofproto_port obtained from
1258 * ofproto_port_dump_next(): that function retains ownership of the data in the
1261 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1263 free(ofproto_port->name);
1264 free(ofproto_port->type);
1267 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1269 * This function provides no status indication. An error status for the entire
1270 * dump operation is provided when it is completed by calling
1271 * ofproto_port_dump_done().
1274 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1275 const struct ofproto *ofproto)
1277 dump->ofproto = ofproto;
1278 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1282 /* Attempts to retrieve another port from 'dump', which must have been created
1283 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1284 * 'port' and returns true. On failure, returns false.
1286 * Failure might indicate an actual error or merely that the last port has been
1287 * dumped. An error status for the entire dump operation is provided when it
1288 * is completed by calling ofproto_port_dump_done().
1290 * The ofproto owns the data stored in 'port'. It will remain valid until at
1291 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1292 * ofproto_port_dump_done(). */
1294 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1295 struct ofproto_port *port)
1297 const struct ofproto *ofproto = dump->ofproto;
1303 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1306 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1312 /* Completes port table dump operation 'dump', which must have been created
1313 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1314 * error-free, otherwise a positive errno value describing the problem. */
1316 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1318 const struct ofproto *ofproto = dump->ofproto;
1320 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1323 return dump->error == EOF ? 0 : dump->error;
1326 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1327 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1328 * is non-null). On failure, returns a positive errno value and sets
1329 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1331 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1332 uint16_t *ofp_portp)
1337 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1339 update_port(ofproto, netdev_get_name(netdev));
1342 *ofp_portp = error ? OFPP_NONE : ofp_port;
1347 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1348 * initializes '*port' appropriately; on failure, returns a positive errno
1351 * The caller owns the data in 'ofproto_port' and must free it with
1352 * ofproto_port_destroy() when it is no longer needed. */
1354 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1355 struct ofproto_port *port)
1359 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1361 memset(port, 0, sizeof *port);
1366 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1367 * Returns 0 if successful, otherwise a positive errno. */
1369 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1371 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1372 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1375 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1376 if (!error && ofport) {
1377 /* 'name' is the netdev's name and update_port() is going to close the
1378 * netdev. Just in case update_port() refers to 'name' after it
1379 * destroys 'ofport', make a copy of it around the update_port()
1381 char *devname = xstrdup(name);
1382 update_port(ofproto, devname);
1388 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1389 * performs the 'n_actions' actions in 'actions'. The new flow will not
1392 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1393 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1394 * controllers; otherwise, it will be hidden.
1396 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1398 * This is a helper function for in-band control and fail-open. */
1400 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1401 const struct ofpact *ofpacts, size_t ofpacts_len)
1403 const struct rule *rule;
1405 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1406 &ofproto->tables[0].cls, cls_rule));
1407 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1408 ofpacts, ofpacts_len)) {
1409 struct ofputil_flow_mod fm;
1411 memset(&fm, 0, sizeof fm);
1413 fm.buffer_id = UINT32_MAX;
1414 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1415 fm.ofpacts_len = ofpacts_len;
1416 add_flow(ofproto, NULL, &fm, NULL);
1421 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1422 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1423 * operation cannot be initiated now but may be retried later.
1425 * This is a helper function for in-band control and fail-open. */
1427 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1429 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1432 /* Searches for a rule with matching criteria exactly equal to 'target' in
1433 * ofproto's table 0 and, if it finds one, deletes it.
1435 * This is a helper function for in-band control and fail-open. */
1437 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1441 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1442 &ofproto->tables[0].cls, target));
1444 /* No such rule -> success. */
1446 } else if (rule->pending) {
1447 /* An operation on the rule is already pending -> failure.
1448 * Caller must retry later if it's important. */
1451 /* Initiate deletion -> success. */
1452 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1453 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1454 oftable_remove_rule(rule);
1455 ofproto->ofproto_class->rule_destruct(rule);
1456 ofopgroup_submit(group);
1462 /* Starts the process of deleting all of the flows from all of ofproto's flow
1463 * tables and then reintroducing the flows required by in-band control and
1464 * fail-open. The process will complete in a later call to ofproto_run(). */
1466 ofproto_flush_flows(struct ofproto *ofproto)
1468 COVERAGE_INC(ofproto_flush);
1469 ofproto->state = S_FLUSH;
1473 reinit_ports(struct ofproto *p)
1475 struct ofproto_port_dump dump;
1476 struct sset devnames;
1477 struct ofport *ofport;
1478 struct ofproto_port ofproto_port;
1479 const char *devname;
1481 COVERAGE_INC(ofproto_reinit_ports);
1483 sset_init(&devnames);
1484 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1485 sset_add(&devnames, netdev_get_name(ofport->netdev));
1487 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1488 sset_add(&devnames, ofproto_port.name);
1491 SSET_FOR_EACH (devname, &devnames) {
1492 update_port(p, devname);
1494 sset_destroy(&devnames);
1497 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1498 * pointer if the netdev cannot be opened. On success, also fills in
1500 static struct netdev *
1501 ofport_open(const struct ofproto *ofproto,
1502 const struct ofproto_port *ofproto_port,
1503 struct ofputil_phy_port *pp)
1505 enum netdev_flags flags;
1506 struct netdev *netdev;
1509 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1511 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1512 "cannot be opened (%s)",
1514 ofproto_port->name, ofproto_port->ofp_port,
1515 ofproto_port->name, strerror(error));
1519 pp->port_no = ofproto_port->ofp_port;
1520 netdev_get_etheraddr(netdev, pp->hw_addr);
1521 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1522 netdev_get_flags(netdev, &flags);
1523 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1524 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1525 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1526 &pp->supported, &pp->peer);
1527 pp->curr_speed = netdev_features_to_bps(pp->curr);
1528 pp->max_speed = netdev_features_to_bps(pp->supported);
1533 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1534 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1537 ofport_equal(const struct ofputil_phy_port *a,
1538 const struct ofputil_phy_port *b)
1540 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1541 && a->state == b->state
1542 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1543 && a->curr == b->curr
1544 && a->advertised == b->advertised
1545 && a->supported == b->supported
1546 && a->peer == b->peer
1547 && a->curr_speed == b->curr_speed
1548 && a->max_speed == b->max_speed);
1551 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1552 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1553 * one with the same name or port number). */
1555 ofport_install(struct ofproto *p,
1556 struct netdev *netdev, const struct ofputil_phy_port *pp)
1558 const char *netdev_name = netdev_get_name(netdev);
1559 struct ofport *ofport;
1562 /* Create ofport. */
1563 ofport = p->ofproto_class->port_alloc();
1568 ofport->ofproto = p;
1569 ofport->netdev = netdev;
1570 ofport->change_seq = netdev_change_seq(netdev);
1572 ofport->ofp_port = pp->port_no;
1574 /* Add port to 'p'. */
1575 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1576 shash_add(&p->port_by_name, netdev_name, ofport);
1578 update_mtu(p, ofport);
1580 /* Let the ofproto_class initialize its private data. */
1581 error = p->ofproto_class->port_construct(ofport);
1585 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1589 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1590 p->name, netdev_name, strerror(error));
1592 ofport_destroy__(ofport);
1594 netdev_close(netdev);
1598 /* Removes 'ofport' from 'p' and destroys it. */
1600 ofport_remove(struct ofport *ofport)
1602 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1604 ofport_destroy(ofport);
1607 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1610 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1612 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1614 ofport_remove(port);
1618 /* Updates 'port' with new 'pp' description.
1620 * Does not handle a name or port number change. The caller must implement
1621 * such a change as a delete followed by an add. */
1623 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1625 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1626 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1627 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1628 port->pp.state = pp->state;
1629 port->pp.curr = pp->curr;
1630 port->pp.advertised = pp->advertised;
1631 port->pp.supported = pp->supported;
1632 port->pp.peer = pp->peer;
1633 port->pp.curr_speed = pp->curr_speed;
1634 port->pp.max_speed = pp->max_speed;
1636 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1639 /* Update OpenFlow 'state' in 'port' and notify controller. */
1641 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1643 if (port->pp.state != state) {
1644 port->pp.state = state;
1645 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1651 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1653 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1655 if (port->ofproto->ofproto_class->set_realdev) {
1656 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1658 if (port->ofproto->ofproto_class->set_stp_port) {
1659 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1661 if (port->ofproto->ofproto_class->set_cfm) {
1662 port->ofproto->ofproto_class->set_cfm(port, NULL);
1664 if (port->ofproto->ofproto_class->bundle_remove) {
1665 port->ofproto->ofproto_class->bundle_remove(port);
1671 ofport_destroy__(struct ofport *port)
1673 struct ofproto *ofproto = port->ofproto;
1674 const char *name = netdev_get_name(port->netdev);
1676 hmap_remove(&ofproto->ports, &port->hmap_node);
1677 shash_delete(&ofproto->port_by_name,
1678 shash_find(&ofproto->port_by_name, name));
1680 netdev_close(port->netdev);
1681 ofproto->ofproto_class->port_dealloc(port);
1685 ofport_destroy(struct ofport *port)
1688 port->ofproto->ofproto_class->port_destruct(port);
1689 ofport_destroy__(port);
1694 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1696 struct ofport *port;
1698 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1699 hash_int(ofp_port, 0), &ofproto->ports) {
1700 if (port->ofp_port == ofp_port) {
1708 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1710 struct ofproto *ofproto = port->ofproto;
1713 if (ofproto->ofproto_class->port_get_stats) {
1714 error = ofproto->ofproto_class->port_get_stats(port, stats);
1723 update_port(struct ofproto *ofproto, const char *name)
1725 struct ofproto_port ofproto_port;
1726 struct ofputil_phy_port pp;
1727 struct netdev *netdev;
1728 struct ofport *port;
1730 COVERAGE_INC(ofproto_update_port);
1732 /* Fetch 'name''s location and properties from the datapath. */
1733 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1734 ? ofport_open(ofproto, &ofproto_port, &pp)
1737 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1738 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1739 struct netdev *old_netdev = port->netdev;
1741 /* 'name' hasn't changed location. Any properties changed? */
1742 if (!ofport_equal(&port->pp, &pp)) {
1743 ofport_modified(port, &pp);
1746 update_mtu(ofproto, port);
1748 /* Install the newly opened netdev in case it has changed.
1749 * Don't close the old netdev yet in case port_modified has to
1750 * remove a retained reference to it.*/
1751 port->netdev = netdev;
1752 port->change_seq = netdev_change_seq(netdev);
1754 if (port->ofproto->ofproto_class->port_modified) {
1755 port->ofproto->ofproto_class->port_modified(port);
1758 netdev_close(old_netdev);
1760 /* If 'port' is nonnull then its name differs from 'name' and thus
1761 * we should delete it. If we think there's a port named 'name'
1762 * then its port number must be wrong now so delete it too. */
1764 ofport_remove(port);
1766 ofport_remove_with_name(ofproto, name);
1767 ofport_install(ofproto, netdev, &pp);
1770 /* Any port named 'name' is gone now. */
1771 ofport_remove_with_name(ofproto, name);
1773 ofproto_port_destroy(&ofproto_port);
1777 init_ports(struct ofproto *p)
1779 struct ofproto_port_dump dump;
1780 struct ofproto_port ofproto_port;
1782 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1783 uint16_t ofp_port = ofproto_port.ofp_port;
1784 if (ofproto_get_port(p, ofp_port)) {
1785 VLOG_WARN_RL(&rl, "%s: ignoring duplicate port %"PRIu16" "
1786 "in datapath", p->name, ofp_port);
1787 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1788 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1789 p->name, ofproto_port.name);
1791 struct ofputil_phy_port pp;
1792 struct netdev *netdev;
1794 netdev = ofport_open(p, &ofproto_port, &pp);
1796 ofport_install(p, netdev, &pp);
1804 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1805 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1807 find_min_mtu(struct ofproto *p)
1809 struct ofport *ofport;
1812 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1813 struct netdev *netdev = ofport->netdev;
1816 /* Skip any internal ports, since that's what we're trying to
1818 if (!strcmp(netdev_get_type(netdev), "internal")) {
1822 if (netdev_get_mtu(netdev, &dev_mtu)) {
1825 if (!mtu || dev_mtu < mtu) {
1830 return mtu ? mtu: ETH_PAYLOAD_MAX;
1833 /* Update MTU of all datapath devices on 'p' to the minimum of the
1834 * non-datapath ports in event of 'port' added or changed. */
1836 update_mtu(struct ofproto *p, struct ofport *port)
1838 struct ofport *ofport;
1839 struct netdev *netdev = port->netdev;
1840 int dev_mtu, old_min;
1842 if (netdev_get_mtu(netdev, &dev_mtu)) {
1846 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1847 if (dev_mtu > p->min_mtu) {
1848 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1849 dev_mtu = p->min_mtu;
1852 port->mtu = dev_mtu;
1856 /* For non-internal port find new min mtu. */
1857 old_min = p->min_mtu;
1858 port->mtu = dev_mtu;
1859 p->min_mtu = find_min_mtu(p);
1860 if (p->min_mtu == old_min) {
1864 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1865 struct netdev *netdev = ofport->netdev;
1867 if (!strcmp(netdev_get_type(netdev), "internal")) {
1868 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1869 ofport->mtu = p->min_mtu;
1876 ofproto_rule_destroy__(struct rule *rule)
1879 free(rule->ofpacts);
1880 rule->ofproto->ofproto_class->rule_dealloc(rule);
1884 /* This function allows an ofproto implementation to destroy any rules that
1885 * remain when its ->destruct() function is called. The caller must have
1886 * already uninitialized any derived members of 'rule' (step 5 described in the
1887 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1888 * This function implements steps 6 and 7.
1890 * This function should only be called from an ofproto implementation's
1891 * ->destruct() function. It is not suitable elsewhere. */
1893 ofproto_rule_destroy(struct rule *rule)
1895 assert(!rule->pending);
1896 oftable_remove_rule(rule);
1897 ofproto_rule_destroy__(rule);
1900 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1901 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1903 ofproto_rule_has_out_port(const struct rule *rule, uint16_t port)
1905 return (port == OFPP_NONE
1906 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
1909 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
1910 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
1912 ofoperation_has_out_port(const struct ofoperation *op, uint16_t out_port)
1914 if (ofproto_rule_has_out_port(op->rule, out_port)) {
1919 case OFOPERATION_ADD:
1920 return op->victim && ofproto_rule_has_out_port(op->victim, out_port);
1922 case OFOPERATION_DELETE:
1925 case OFOPERATION_MODIFY:
1926 return ofpacts_output_to_port(op->ofpacts, op->ofpacts_len, out_port);
1932 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1933 * statistics appropriately. 'packet' must have at least sizeof(struct
1934 * ofp_packet_in) bytes of headroom.
1936 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1937 * with statistics for 'packet' either way.
1939 * Takes ownership of 'packet'. */
1941 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1945 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1947 flow_extract(packet, 0, 0, in_port, &flow);
1948 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1951 /* Returns true if 'rule' should be hidden from the controller.
1953 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1954 * (e.g. by in-band control) and are intentionally hidden from the
1957 ofproto_rule_is_hidden(const struct rule *rule)
1959 return rule->cr.priority > UINT16_MAX;
1962 static enum oftable_flags
1963 rule_get_flags(const struct rule *rule)
1965 return rule->ofproto->tables[rule->table_id].flags;
1969 rule_is_modifiable(const struct rule *rule)
1971 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1975 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1977 ofconn_send_reply(ofconn, make_echo_reply(oh));
1982 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1984 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1985 struct ofputil_switch_features features;
1986 struct ofport *port;
1990 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1992 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1994 features.datapath_id = ofproto->datapath_id;
1995 features.n_buffers = pktbuf_capacity();
1996 features.n_tables = ofproto->n_tables;
1997 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1998 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2000 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2003 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2005 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2006 ofputil_put_switch_features_port(&port->pp, b);
2009 ofconn_send_reply(ofconn, b);
2014 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2016 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2017 struct ofp_switch_config *osc;
2018 enum ofp_config_flags flags;
2022 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2023 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2024 flags = ofproto->frag_handling;
2025 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
2026 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2028 osc->flags = htons(flags);
2029 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2030 ofconn_send_reply(ofconn, buf);
2036 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2038 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2039 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2040 uint16_t flags = ntohs(osc->flags);
2042 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2043 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2044 enum ofp_config_flags cur = ofproto->frag_handling;
2045 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2047 assert((cur & OFPC_FRAG_MASK) == cur);
2049 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2050 ofproto->frag_handling = next;
2052 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2054 ofputil_frag_handling_to_string(next));
2058 ofconn_set_invalid_ttl_to_controller(ofconn,
2059 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2061 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2066 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2067 * error message code for the caller to propagate upward. Otherwise, returns
2070 * The log message mentions 'msg_type'. */
2072 reject_slave_controller(struct ofconn *ofconn)
2074 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2075 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2076 return OFPERR_OFPBRC_EPERM;
2083 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2085 struct ofproto *p = ofconn_get_ofproto(ofconn);
2086 struct ofputil_packet_out po;
2087 struct ofpbuf *payload;
2088 uint64_t ofpacts_stub[1024 / 8];
2089 struct ofpbuf ofpacts;
2093 COVERAGE_INC(ofproto_packet_out);
2095 error = reject_slave_controller(ofconn);
2100 /* Decode message. */
2101 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2102 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2104 goto exit_free_ofpacts;
2108 if (po.buffer_id != UINT32_MAX) {
2109 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2110 if (error || !payload) {
2111 goto exit_free_ofpacts;
2114 payload = xmalloc(sizeof *payload);
2115 ofpbuf_use_const(payload, po.packet, po.packet_len);
2118 /* Send out packet. */
2119 flow_extract(payload, 0, 0, po.in_port, &flow);
2120 error = p->ofproto_class->packet_out(p, payload, &flow,
2121 po.ofpacts, po.ofpacts_len);
2122 ofpbuf_delete(payload);
2125 ofpbuf_uninit(&ofpacts);
2131 update_port_config(struct ofport *port,
2132 enum ofputil_port_config config,
2133 enum ofputil_port_config mask)
2135 enum ofputil_port_config old_config = port->pp.config;
2136 enum ofputil_port_config toggle;
2138 toggle = (config ^ port->pp.config) & mask;
2139 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2140 if (config & OFPUTIL_PC_PORT_DOWN) {
2141 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2143 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2145 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2148 port->pp.config ^= toggle;
2149 if (port->pp.config != old_config) {
2150 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2155 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2157 struct ofproto *p = ofconn_get_ofproto(ofconn);
2158 struct ofputil_port_mod pm;
2159 struct ofport *port;
2162 error = reject_slave_controller(ofconn);
2167 error = ofputil_decode_port_mod(oh, &pm);
2172 port = ofproto_get_port(p, pm.port_no);
2174 return OFPERR_OFPPMFC_BAD_PORT;
2175 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2176 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2178 update_port_config(port, pm.config, pm.mask);
2180 netdev_set_advertisements(port->netdev, pm.advertise);
2187 handle_desc_stats_request(struct ofconn *ofconn,
2188 const struct ofp_header *request)
2190 struct ofproto *p = ofconn_get_ofproto(ofconn);
2191 struct ofp_desc_stats *ods;
2194 msg = ofpraw_alloc_stats_reply(request, 0);
2195 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2196 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2197 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2198 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2199 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2200 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2201 ofconn_send_reply(ofconn, msg);
2207 handle_table_stats_request(struct ofconn *ofconn,
2208 const struct ofp_header *request)
2210 struct ofproto *p = ofconn_get_ofproto(ofconn);
2211 struct ofp10_table_stats *ots;
2215 msg = ofpraw_alloc_stats_reply(request, sizeof *ots * p->n_tables);
2216 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2217 for (i = 0; i < p->n_tables; i++) {
2218 ots[i].table_id = i;
2219 sprintf(ots[i].name, "table%zu", i);
2220 ots[i].wildcards = htonl(OFPFW10_ALL);
2221 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2222 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2225 p->ofproto_class->get_tables(p, ots);
2227 for (i = 0; i < p->n_tables; i++) {
2228 const struct oftable *table = &p->tables[i];
2231 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2234 if (table->max_flows < ntohl(ots[i].max_entries)) {
2235 ots[i].max_entries = htonl(table->max_flows);
2239 ofconn_send_reply(ofconn, msg);
2244 append_port_stat(struct ofport *port, struct list *replies)
2246 struct netdev_stats stats;
2247 struct ofp10_port_stats *ops;
2249 /* Intentionally ignore return value, since errors will set
2250 * 'stats' to all-1s, which is correct for OpenFlow, and
2251 * netdev_get_stats() will log errors. */
2252 ofproto_port_get_stats(port, &stats);
2254 ops = ofpmp_append(replies, sizeof *ops);
2255 ops->port_no = htons(port->pp.port_no);
2256 memset(ops->pad, 0, sizeof ops->pad);
2257 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2258 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2259 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2260 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2261 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2262 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2263 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2264 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2265 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2266 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2267 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2268 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2272 handle_port_stats_request(struct ofconn *ofconn,
2273 const struct ofp_header *request)
2275 struct ofproto *p = ofconn_get_ofproto(ofconn);
2276 const struct ofp10_port_stats_request *psr = ofpmsg_body(request);
2277 struct ofport *port;
2278 struct list replies;
2280 ofpmp_init(&replies, request);
2281 if (psr->port_no != htons(OFPP_NONE)) {
2282 port = ofproto_get_port(p, ntohs(psr->port_no));
2284 append_port_stat(port, &replies);
2287 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2288 append_port_stat(port, &replies);
2292 ofconn_send_replies(ofconn, &replies);
2297 handle_port_desc_stats_request(struct ofconn *ofconn,
2298 const struct ofp_header *request)
2300 struct ofproto *p = ofconn_get_ofproto(ofconn);
2301 enum ofp_version version;
2302 struct ofport *port;
2303 struct list replies;
2305 ofpmp_init(&replies, request);
2307 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
2308 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2309 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
2312 ofconn_send_replies(ofconn, &replies);
2317 calc_flow_duration__(long long int start, long long int now,
2318 uint32_t *sec, uint32_t *nsec)
2320 long long int msecs = now - start;
2321 *sec = msecs / 1000;
2322 *nsec = (msecs % 1000) * (1000 * 1000);
2325 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2326 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2328 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2330 return (table_id == 0xff || table_id < ofproto->n_tables
2332 : OFPERR_NXBRC_BAD_TABLE_ID);
2336 static struct oftable *
2337 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
2339 struct oftable *table;
2341 for (table = &ofproto->tables[table_id];
2342 table < &ofproto->tables[ofproto->n_tables];
2344 if (!(table->flags & OFTABLE_HIDDEN)) {
2352 static struct oftable *
2353 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
2355 if (table_id == 0xff) {
2356 return next_visible_table(ofproto, 0);
2357 } else if (table_id < ofproto->n_tables) {
2358 return &ofproto->tables[table_id];
2364 static struct oftable *
2365 next_matching_table(const struct ofproto *ofproto,
2366 const struct oftable *table, uint8_t table_id)
2368 return (table_id == 0xff
2369 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2373 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2375 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2376 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2378 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2379 * only once, for that table. (This can be used to access tables marked
2382 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2383 * entered at all. (Perhaps you should have validated TABLE_ID with
2384 * check_table_id().)
2386 * All parameters are evaluated multiple times.
2388 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2389 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2391 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2393 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2394 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2395 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2398 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2399 * to 'out_port' are included.
2401 * Hidden rules are always omitted.
2403 * Returns 0 on success, otherwise an OpenFlow error code. */
2405 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2406 const struct cls_rule *match,
2407 ovs_be64 cookie, ovs_be64 cookie_mask,
2408 uint16_t out_port, struct list *rules)
2410 struct oftable *table;
2413 error = check_table_id(ofproto, table_id);
2419 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2420 struct cls_cursor cursor;
2423 cls_cursor_init(&cursor, &table->cls, match);
2424 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2425 if (rule->pending) {
2426 return OFPROTO_POSTPONE;
2428 if (!ofproto_rule_is_hidden(rule)
2429 && ofproto_rule_has_out_port(rule, out_port)
2430 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2431 list_push_back(rules, &rule->ofproto_node);
2438 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2439 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2440 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2443 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2444 * to 'out_port' are included.
2446 * Hidden rules are always omitted.
2448 * Returns 0 on success, otherwise an OpenFlow error code. */
2450 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2451 const struct cls_rule *match,
2452 ovs_be64 cookie, ovs_be64 cookie_mask,
2453 uint16_t out_port, struct list *rules)
2455 struct oftable *table;
2458 error = check_table_id(ofproto, table_id);
2464 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2467 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2470 if (rule->pending) {
2471 return OFPROTO_POSTPONE;
2473 if (!ofproto_rule_is_hidden(rule)
2474 && ofproto_rule_has_out_port(rule, out_port)
2475 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2476 list_push_back(rules, &rule->ofproto_node);
2483 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2484 * forced into the range of a uint16_t. */
2486 age_secs(long long int age_ms)
2488 return (age_ms < 0 ? 0
2489 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2490 : (unsigned int) age_ms / 1000);
2494 handle_flow_stats_request(struct ofconn *ofconn,
2495 const struct ofp_header *request)
2497 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2498 struct ofputil_flow_stats_request fsr;
2499 struct list replies;
2504 error = ofputil_decode_flow_stats_request(&fsr, request);
2509 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2510 fsr.cookie, fsr.cookie_mask,
2511 fsr.out_port, &rules);
2516 ofpmp_init(&replies, request);
2517 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2518 long long int now = time_msec();
2519 struct ofputil_flow_stats fs;
2522 fs.cookie = rule->flow_cookie;
2523 fs.table_id = rule->table_id;
2524 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2526 fs.idle_timeout = rule->idle_timeout;
2527 fs.hard_timeout = rule->hard_timeout;
2528 fs.idle_age = age_secs(now - rule->used);
2529 fs.hard_age = age_secs(now - rule->modified);
2530 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2532 fs.ofpacts = rule->ofpacts;
2533 fs.ofpacts_len = rule->ofpacts_len;
2534 ofputil_append_flow_stats_reply(&fs, &replies);
2536 ofconn_send_replies(ofconn, &replies);
2542 flow_stats_ds(struct rule *rule, struct ds *results)
2544 uint64_t packet_count, byte_count;
2546 rule->ofproto->ofproto_class->rule_get_stats(rule,
2547 &packet_count, &byte_count);
2549 if (rule->table_id != 0) {
2550 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2552 ds_put_format(results, "duration=%llds, ",
2553 (time_msec() - rule->created) / 1000);
2554 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2555 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2556 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2557 cls_rule_format(&rule->cr, results);
2558 ds_put_char(results, ',');
2559 if (rule->ofpacts_len > 0) {
2560 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2562 ds_put_cstr(results, "drop");
2564 ds_put_cstr(results, "\n");
2567 /* Adds a pretty-printed description of all flows to 'results', including
2568 * hidden flows (e.g., set up by in-band control). */
2570 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2572 struct oftable *table;
2574 OFPROTO_FOR_EACH_TABLE (table, p) {
2575 struct cls_cursor cursor;
2578 cls_cursor_init(&cursor, &table->cls, NULL);
2579 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2580 flow_stats_ds(rule, results);
2585 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2586 * '*engine_type' and '*engine_id', respectively. */
2588 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2589 uint8_t *engine_type, uint8_t *engine_id)
2591 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2594 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2595 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2596 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2597 * and -1 if CFM is not enabled on 'ofp_port'. */
2599 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2601 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2602 return (ofport && ofproto->ofproto_class->get_cfm_fault
2603 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2607 /* Checks the operational status reported by the remote CFM endpoint of
2608 * 'ofp_port' Returns 1 if operationally up, 0 if operationally down, and -1
2609 * if CFM is not enabled on 'ofp_port' or does not support operational status.
2612 ofproto_port_get_cfm_opup(const struct ofproto *ofproto, uint16_t ofp_port)
2614 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2615 return (ofport && ofproto->ofproto_class->get_cfm_opup
2616 ? ofproto->ofproto_class->get_cfm_opup(ofport)
2620 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2621 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2622 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2623 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2625 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2626 uint16_t ofp_port, const uint64_t **rmps,
2629 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2633 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2634 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2639 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2640 * integer value between 0 and 100 to indicate the health of the port as a
2641 * percentage which is the average of cfm health of all the remote_mpids or
2642 * returns -1 if CFM is not enabled on 'ofport'. */
2644 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2646 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2647 return (ofport && ofproto->ofproto_class->get_cfm_health
2648 ? ofproto->ofproto_class->get_cfm_health(ofport)
2653 handle_aggregate_stats_request(struct ofconn *ofconn,
2654 const struct ofp_header *oh)
2656 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2657 struct ofputil_flow_stats_request request;
2658 struct ofputil_aggregate_stats stats;
2659 bool unknown_packets, unknown_bytes;
2660 struct ofpbuf *reply;
2665 error = ofputil_decode_flow_stats_request(&request, oh);
2670 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2671 request.cookie, request.cookie_mask,
2672 request.out_port, &rules);
2677 memset(&stats, 0, sizeof stats);
2678 unknown_packets = unknown_bytes = false;
2679 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2680 uint64_t packet_count;
2681 uint64_t byte_count;
2683 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2686 if (packet_count == UINT64_MAX) {
2687 unknown_packets = true;
2689 stats.packet_count += packet_count;
2692 if (byte_count == UINT64_MAX) {
2693 unknown_bytes = true;
2695 stats.byte_count += byte_count;
2700 if (unknown_packets) {
2701 stats.packet_count = UINT64_MAX;
2703 if (unknown_bytes) {
2704 stats.byte_count = UINT64_MAX;
2707 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
2708 ofconn_send_reply(ofconn, reply);
2713 struct queue_stats_cbdata {
2714 struct ofport *ofport;
2715 struct list replies;
2719 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2720 const struct netdev_queue_stats *stats)
2722 struct ofp10_queue_stats *reply;
2724 reply = ofpmp_append(&cbdata->replies, sizeof *reply);
2725 reply->port_no = htons(cbdata->ofport->pp.port_no);
2726 memset(reply->pad, 0, sizeof reply->pad);
2727 reply->queue_id = htonl(queue_id);
2728 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2729 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2730 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2734 handle_queue_stats_dump_cb(uint32_t queue_id,
2735 struct netdev_queue_stats *stats,
2738 struct queue_stats_cbdata *cbdata = cbdata_;
2740 put_queue_stats(cbdata, queue_id, stats);
2744 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2745 struct queue_stats_cbdata *cbdata)
2747 cbdata->ofport = port;
2748 if (queue_id == OFPQ_ALL) {
2749 netdev_dump_queue_stats(port->netdev,
2750 handle_queue_stats_dump_cb, cbdata);
2752 struct netdev_queue_stats stats;
2754 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2755 put_queue_stats(cbdata, queue_id, &stats);
2757 return OFPERR_OFPQOFC_BAD_QUEUE;
2764 handle_queue_stats_request(struct ofconn *ofconn,
2765 const struct ofp_header *rq)
2767 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2768 const struct ofp10_queue_stats_request *qsr = ofpmsg_body(rq);
2769 struct queue_stats_cbdata cbdata;
2770 unsigned int port_no;
2771 struct ofport *port;
2775 COVERAGE_INC(ofproto_queue_req);
2777 ofpmp_init(&cbdata.replies, rq);
2779 port_no = ntohs(qsr->port_no);
2780 queue_id = ntohl(qsr->queue_id);
2781 if (port_no == OFPP_ALL) {
2782 error = OFPERR_OFPQOFC_BAD_QUEUE;
2783 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2784 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2789 port = ofproto_get_port(ofproto, port_no);
2791 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2792 : OFPERR_OFPQOFC_BAD_PORT);
2795 ofconn_send_replies(ofconn, &cbdata.replies);
2797 ofpbuf_list_delete(&cbdata.replies);
2804 is_flow_deletion_pending(const struct ofproto *ofproto,
2805 const struct cls_rule *cls_rule,
2808 if (!hmap_is_empty(&ofproto->deletions)) {
2809 struct ofoperation *op;
2811 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2812 cls_rule_hash(cls_rule, table_id),
2813 &ofproto->deletions) {
2814 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2823 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2824 * in which no matching flow already exists in the flow table.
2826 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2827 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2828 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2829 * initiated now but may be retried later.
2831 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2832 * ownership remains with the caller.
2834 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2837 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2838 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2840 struct oftable *table;
2841 struct ofopgroup *group;
2842 struct rule *victim;
2846 error = check_table_id(ofproto, fm->table_id);
2852 if (fm->table_id == 0xff) {
2854 if (ofproto->ofproto_class->rule_choose_table) {
2855 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2860 assert(table_id < ofproto->n_tables);
2861 table = &ofproto->tables[table_id];
2863 table = &ofproto->tables[0];
2865 } else if (fm->table_id < ofproto->n_tables) {
2866 table = &ofproto->tables[fm->table_id];
2868 return OFPERR_NXFMFC_BAD_TABLE_ID;
2871 if (table->flags & OFTABLE_READONLY) {
2872 return OFPERR_OFPBRC_EPERM;
2875 /* Check for overlap, if requested. */
2876 if (fm->flags & OFPFF_CHECK_OVERLAP
2877 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2878 return OFPERR_OFPFMFC_OVERLAP;
2881 /* Serialize against pending deletion. */
2882 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2883 return OFPROTO_POSTPONE;
2886 /* Allocate new rule. */
2887 rule = ofproto->ofproto_class->rule_alloc();
2889 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2890 ofproto->name, strerror(error));
2893 rule->ofproto = ofproto;
2895 rule->pending = NULL;
2896 rule->flow_cookie = fm->new_cookie;
2897 rule->created = rule->modified = rule->used = time_msec();
2898 rule->idle_timeout = fm->idle_timeout;
2899 rule->hard_timeout = fm->hard_timeout;
2900 rule->table_id = table - ofproto->tables;
2901 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2902 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2903 rule->ofpacts_len = fm->ofpacts_len;
2904 rule->evictable = true;
2905 rule->eviction_group = NULL;
2906 rule->monitor_flags = 0;
2907 rule->add_seqno = 0;
2908 rule->modify_seqno = 0;
2910 /* Insert new rule. */
2911 victim = oftable_replace_rule(rule);
2912 if (victim && !rule_is_modifiable(victim)) {
2913 error = OFPERR_OFPBRC_EPERM;
2914 } else if (victim && victim->pending) {
2915 error = OFPROTO_POSTPONE;
2917 struct ofoperation *op;
2920 if (classifier_count(&table->cls) > table->max_flows) {
2923 was_evictable = rule->evictable;
2924 rule->evictable = false;
2925 evict = choose_rule_to_evict(table);
2926 rule->evictable = was_evictable;
2929 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2931 } else if (evict->pending) {
2932 error = OFPROTO_POSTPONE;
2939 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2940 op = ofoperation_create(group, rule, OFOPERATION_ADD, 0);
2941 op->victim = victim;
2943 error = ofproto->ofproto_class->rule_construct(rule);
2945 op->group->n_running--;
2946 ofoperation_destroy(rule->pending);
2948 delete_flow__(evict, group);
2950 ofopgroup_submit(group);
2954 /* Back out if an error occurred. */
2956 oftable_substitute_rule(rule, victim);
2957 ofproto_rule_destroy__(rule);
2962 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2964 /* Modifies the rules listed in 'rules', changing their actions to match those
2967 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2970 * Returns 0 on success, otherwise an OpenFlow error code. */
2972 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2973 const struct ofputil_flow_mod *fm,
2974 const struct ofp_header *request, struct list *rules)
2976 struct ofopgroup *group;
2980 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2981 error = OFPERR_OFPBRC_EPERM;
2982 LIST_FOR_EACH (rule, ofproto_node, rules) {
2983 struct ofoperation *op;
2984 bool actions_changed;
2985 ovs_be64 new_cookie;
2987 if (rule_is_modifiable(rule)) {
2988 /* At least one rule is modifiable, don't report EPERM error. */
2994 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2995 rule->ofpacts, rule->ofpacts_len);
2996 new_cookie = (fm->new_cookie != htonll(UINT64_MAX)
2998 : rule->flow_cookie);
2999 if (!actions_changed && new_cookie == rule->flow_cookie) {
3000 /* No change at all. */
3004 op = ofoperation_create(group, rule, OFOPERATION_MODIFY, 0);
3005 rule->flow_cookie = new_cookie;
3006 if (actions_changed) {
3007 op->ofpacts = rule->ofpacts;
3008 op->ofpacts_len = rule->ofpacts_len;
3009 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
3010 rule->ofpacts_len = fm->ofpacts_len;
3011 rule->ofproto->ofproto_class->rule_modify_actions(rule);
3013 ofoperation_complete(op, 0);
3016 ofopgroup_submit(group);
3021 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
3024 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3027 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3028 const struct ofputil_flow_mod *fm,
3029 const struct ofp_header *request)
3034 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3035 fm->cookie, fm->cookie_mask,
3039 } else if (list_is_empty(&rules)) {
3040 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
3042 return modify_flows__(ofproto, ofconn, fm, request, &rules);
3046 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3049 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
3052 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3053 const struct ofputil_flow_mod *fm,
3054 const struct ofp_header *request)
3059 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3060 fm->cookie, fm->cookie_mask,
3065 } else if (list_is_empty(&rules)) {
3066 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
3068 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
3069 fm, request, &rules)
3074 /* OFPFC_DELETE implementation. */
3077 delete_flow__(struct rule *rule, struct ofopgroup *group)
3079 struct ofproto *ofproto = rule->ofproto;
3081 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3083 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
3084 oftable_remove_rule(rule);
3085 ofproto->ofproto_class->rule_destruct(rule);
3088 /* Deletes the rules listed in 'rules'.
3090 * Returns 0 on success, otherwise an OpenFlow error code. */
3092 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3093 const struct ofp_header *request, struct list *rules)
3095 struct rule *rule, *next;
3096 struct ofopgroup *group;
3098 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3099 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3100 delete_flow__(rule, group);
3102 ofopgroup_submit(group);
3107 /* Implements OFPFC_DELETE. */
3109 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3110 const struct ofputil_flow_mod *fm,
3111 const struct ofp_header *request)
3116 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3117 fm->cookie, fm->cookie_mask,
3118 fm->out_port, &rules);
3119 return (error ? error
3120 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3125 /* Implements OFPFC_DELETE_STRICT. */
3127 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3128 const struct ofputil_flow_mod *fm,
3129 const struct ofp_header *request)
3134 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3135 fm->cookie, fm->cookie_mask,
3136 fm->out_port, &rules);
3137 return (error ? error
3138 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3144 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3146 struct ofputil_flow_removed fr;
3148 if (ofproto_rule_is_hidden(rule) || !rule->send_flow_removed) {
3153 fr.cookie = rule->flow_cookie;
3155 calc_flow_duration__(rule->created, time_msec(),
3156 &fr.duration_sec, &fr.duration_nsec);
3157 fr.idle_timeout = rule->idle_timeout;
3158 fr.hard_timeout = rule->hard_timeout;
3159 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3162 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3166 ofproto_rule_update_used(struct rule *rule, long long int used)
3168 if (used > rule->used) {
3169 struct eviction_group *evg = rule->eviction_group;
3173 heap_change(&evg->rules, &rule->evg_node,
3174 rule_eviction_priority(rule));
3179 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3180 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3183 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
3186 * ofproto implementation ->run() functions should use this function to expire
3187 * OpenFlow flows. */
3189 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3191 struct ofproto *ofproto = rule->ofproto;
3192 struct ofopgroup *group;
3194 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3196 ofproto_rule_send_removed(rule, reason);
3198 group = ofopgroup_create_unattached(ofproto);
3199 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
3200 oftable_remove_rule(rule);
3201 ofproto->ofproto_class->rule_destruct(rule);
3202 ofopgroup_submit(group);
3206 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3208 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3209 struct ofputil_flow_mod fm;
3210 uint64_t ofpacts_stub[1024 / 8];
3211 struct ofpbuf ofpacts;
3215 error = reject_slave_controller(ofconn);
3220 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3221 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3224 goto exit_free_ofpacts;
3227 if (fm.flags & OFPFF10_EMERG) {
3228 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3229 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3230 /* We do not support the emergency flow cache. It will hopefully get
3231 * dropped from OpenFlow in the near future. There is no good error
3232 * code, so just state that the flow table is full. */
3233 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
3235 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3238 goto exit_free_ofpacts;
3241 /* Record the operation for logging a summary report. */
3242 switch (fm.command) {
3248 case OFPFC_MODIFY_STRICT:
3249 ofproto->n_modify++;
3253 case OFPFC_DELETE_STRICT:
3254 ofproto->n_delete++;
3259 if (ofproto->next_op_report == LLONG_MAX) {
3260 ofproto->first_op = now;
3261 ofproto->next_op_report = MAX(now + 10 * 1000,
3262 ofproto->op_backoff);
3263 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3265 ofproto->last_op = now;
3268 ofpbuf_uninit(&ofpacts);
3274 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3275 const struct ofputil_flow_mod *fm,
3276 const struct ofp_header *oh)
3278 if (ofproto->n_pending >= 50) {
3279 assert(!list_is_empty(&ofproto->pending));
3280 return OFPROTO_POSTPONE;
3283 switch (fm->command) {
3285 return add_flow(ofproto, ofconn, fm, oh);
3288 return modify_flows_loose(ofproto, ofconn, fm, oh);
3290 case OFPFC_MODIFY_STRICT:
3291 return modify_flow_strict(ofproto, ofconn, fm, oh);
3294 return delete_flows_loose(ofproto, ofconn, fm, oh);
3296 case OFPFC_DELETE_STRICT:
3297 return delete_flow_strict(ofproto, ofconn, fm, oh);
3300 if (fm->command > 0xff) {
3301 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3302 "flow_mod_table_id extension is not enabled",
3305 return OFPERR_OFPFMFC_BAD_COMMAND;
3310 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3312 const struct nx_role_request *nrr = ofpmsg_body(oh);
3313 struct nx_role_request *reply;
3317 role = ntohl(nrr->role);
3318 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3319 && role != NX_ROLE_SLAVE) {
3320 return OFPERR_OFPRRFC_BAD_ROLE;
3323 if (ofconn_get_role(ofconn) != role
3324 && ofconn_has_pending_opgroups(ofconn)) {
3325 return OFPROTO_POSTPONE;
3328 ofconn_set_role(ofconn, role);
3330 buf = ofpraw_alloc_reply(OFPRAW_NXT_ROLE_REPLY, oh, 0);
3331 reply = ofpbuf_put_zeros(buf, sizeof *reply);
3332 reply->role = htonl(role);
3333 ofconn_send_reply(ofconn, buf);
3339 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3340 const struct ofp_header *oh)
3342 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
3343 enum ofputil_protocol cur, next;
3345 cur = ofconn_get_protocol(ofconn);
3346 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3347 ofconn_set_protocol(ofconn, next);
3353 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3355 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
3356 enum ofputil_protocol cur, next;
3357 enum ofputil_protocol next_base;
3359 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3361 return OFPERR_OFPBRC_EPERM;
3364 cur = ofconn_get_protocol(ofconn);
3365 next = ofputil_protocol_set_base(cur, next_base);
3366 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3367 /* Avoid sending async messages in surprising protocol. */
3368 return OFPROTO_POSTPONE;
3371 ofconn_set_protocol(ofconn, next);
3376 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3377 const struct ofp_header *oh)
3379 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
3382 format = ntohl(msg->format);
3383 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3384 return OFPERR_OFPBRC_EPERM;
3387 if (format != ofconn_get_packet_in_format(ofconn)
3388 && ofconn_has_pending_opgroups(ofconn)) {
3389 /* Avoid sending async message in surprsing packet in format. */
3390 return OFPROTO_POSTPONE;
3393 ofconn_set_packet_in_format(ofconn, format);
3398 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3400 const struct nx_async_config *msg = ofpmsg_body(oh);
3401 uint32_t master[OAM_N_TYPES];
3402 uint32_t slave[OAM_N_TYPES];
3404 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3405 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3406 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3408 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3409 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3410 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3412 ofconn_set_async_config(ofconn, master, slave);
3413 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3414 !ofconn_get_miss_send_len(ofconn)) {
3415 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3422 handle_nxt_set_controller_id(struct ofconn *ofconn,
3423 const struct ofp_header *oh)
3425 const struct nx_controller_id *nci = ofpmsg_body(oh);
3427 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3428 return OFPERR_NXBRC_MUST_BE_ZERO;
3431 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3436 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3440 if (ofconn_has_pending_opgroups(ofconn)) {
3441 return OFPROTO_POSTPONE;
3444 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
3445 ? OFPRAW_OFPT10_BARRIER_REPLY
3446 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
3447 ofconn_send_reply(ofconn, buf);
3452 ofproto_compose_flow_refresh_update(const struct rule *rule,
3453 enum nx_flow_monitor_flags flags,
3456 struct ofoperation *op = rule->pending;
3457 struct ofputil_flow_update fu;
3459 if (op && op->type == OFOPERATION_ADD && !op->victim) {
3460 /* We'll report the final flow when the operation completes. Reporting
3461 * it now would cause a duplicate report later. */
3465 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
3466 ? NXFME_ADDED : NXFME_MODIFIED);
3468 fu.idle_timeout = rule->idle_timeout;
3469 fu.hard_timeout = rule->hard_timeout;
3470 fu.table_id = rule->table_id;
3471 fu.cookie = rule->flow_cookie;
3472 fu.match = CONST_CAST(struct cls_rule *, &rule->cr);
3473 if (!(flags & NXFMF_ACTIONS)) {
3477 fu.ofpacts = rule->ofpacts;
3478 fu.ofpacts_len = rule->ofpacts_len;
3480 /* An operation is in progress. Use the previous version of the flow's
3481 * actions, so that when the operation commits we report the change. */
3483 case OFOPERATION_ADD:
3484 /* We already verified that there was a victim. */
3485 fu.ofpacts = op->victim->ofpacts;
3486 fu.ofpacts_len = op->victim->ofpacts_len;
3489 case OFOPERATION_MODIFY:
3491 fu.ofpacts = op->ofpacts;
3492 fu.ofpacts_len = op->ofpacts_len;
3494 fu.ofpacts = rule->ofpacts;
3495 fu.ofpacts_len = rule->ofpacts_len;
3499 case OFOPERATION_DELETE:
3500 fu.ofpacts = rule->ofpacts;
3501 fu.ofpacts_len = rule->ofpacts_len;
3509 if (list_is_empty(msgs)) {
3510 ofputil_start_flow_update(msgs);
3512 ofputil_append_flow_update(&fu, msgs);
3516 ofmonitor_compose_refresh_updates(struct list *rules, struct list *msgs)
3520 LIST_FOR_EACH (rule, ofproto_node, rules) {
3521 enum nx_flow_monitor_flags flags = rule->monitor_flags;
3522 rule->monitor_flags = 0;
3524 ofproto_compose_flow_refresh_update(rule, flags, msgs);
3529 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
3530 struct rule *rule, uint64_t seqno,
3533 enum nx_flow_monitor_flags update;
3535 if (ofproto_rule_is_hidden(rule)) {
3540 ? ofoperation_has_out_port(rule->pending, m->out_port)
3541 : ofproto_rule_has_out_port(rule, m->out_port))) {
3546 if (rule->add_seqno > seqno) {
3547 update = NXFMF_ADD | NXFMF_MODIFY;
3548 } else if (rule->modify_seqno > seqno) {
3549 update = NXFMF_MODIFY;
3554 if (!(m->flags & update)) {
3558 update = NXFMF_INITIAL;
3561 if (!rule->monitor_flags) {
3562 list_push_back(rules, &rule->ofproto_node);
3564 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
3568 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
3572 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
3573 const struct ofoperation *op;
3574 const struct oftable *table;
3576 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
3577 struct cls_cursor cursor;
3580 cls_cursor_init(&cursor, &table->cls, &m->match);
3581 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3582 assert(!rule->pending); /* XXX */
3583 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3587 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
3588 struct rule *rule = op->rule;
3590 if (((m->table_id == 0xff
3591 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
3592 : m->table_id == rule->table_id))
3593 && cls_rule_is_loose_match(&rule->cr, &m->match)) {
3594 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
3600 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
3603 if (m->flags & NXFMF_INITIAL) {
3604 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
3609 ofmonitor_collect_resume_rules(struct ofmonitor *m,
3610 uint64_t seqno, struct list *rules)
3612 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
3616 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
3618 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3619 struct ofmonitor **monitors;
3620 size_t n_monitors, allocated_monitors;
3621 struct list replies;
3628 ofpbuf_use_const(&b, oh, ntohs(oh->length));
3630 n_monitors = allocated_monitors = 0;
3632 struct ofputil_flow_monitor_request request;
3633 struct ofmonitor *m;
3636 retval = ofputil_decode_flow_monitor_request(&request, &b);
3637 if (retval == EOF) {
3639 } else if (retval) {
3644 if (request.table_id != 0xff
3645 && request.table_id >= ofproto->n_tables) {
3646 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3650 error = ofmonitor_create(&request, ofconn, &m);
3655 if (n_monitors >= allocated_monitors) {
3656 monitors = x2nrealloc(monitors, &allocated_monitors,
3659 monitors[n_monitors++] = m;
3663 for (i = 0; i < n_monitors; i++) {
3664 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
3667 ofpmp_init(&replies, oh);
3668 ofmonitor_compose_refresh_updates(&rules, &replies);
3669 ofconn_send_replies(ofconn, &replies);
3676 for (i = 0; i < n_monitors; i++) {
3677 ofmonitor_destroy(monitors[i]);
3684 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
3686 struct ofmonitor *m;
3689 id = ofputil_decode_flow_monitor_cancel(oh);
3690 m = ofmonitor_lookup(ofconn, id);
3692 return OFPERR_NXBRC_FM_BAD_ID;
3695 ofmonitor_destroy(m);
3700 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3702 const struct ofp_header *oh = msg->data;
3706 error = ofptype_decode(&type, oh);
3712 /* OpenFlow requests. */
3713 case OFPTYPE_ECHO_REQUEST:
3714 return handle_echo_request(ofconn, oh);
3716 case OFPTYPE_FEATURES_REQUEST:
3717 return handle_features_request(ofconn, oh);
3719 case OFPTYPE_GET_CONFIG_REQUEST:
3720 return handle_get_config_request(ofconn, oh);
3722 case OFPTYPE_SET_CONFIG:
3723 return handle_set_config(ofconn, oh);
3725 case OFPTYPE_PACKET_OUT:
3726 return handle_packet_out(ofconn, oh);
3728 case OFPTYPE_PORT_MOD:
3729 return handle_port_mod(ofconn, oh);
3731 case OFPTYPE_FLOW_MOD:
3732 return handle_flow_mod(ofconn, oh);
3734 case OFPTYPE_BARRIER_REQUEST:
3735 return handle_barrier_request(ofconn, oh);
3737 /* OpenFlow replies. */
3738 case OFPTYPE_ECHO_REPLY:
3741 /* Nicira extension requests. */
3742 case OFPTYPE_ROLE_REQUEST:
3743 return handle_role_request(ofconn, oh);
3745 case OFPTYPE_FLOW_MOD_TABLE_ID:
3746 return handle_nxt_flow_mod_table_id(ofconn, oh);
3748 case OFPTYPE_SET_FLOW_FORMAT:
3749 return handle_nxt_set_flow_format(ofconn, oh);
3751 case OFPTYPE_SET_PACKET_IN_FORMAT:
3752 return handle_nxt_set_packet_in_format(ofconn, oh);
3754 case OFPTYPE_SET_CONTROLLER_ID:
3755 return handle_nxt_set_controller_id(ofconn, oh);
3757 case OFPTYPE_FLOW_AGE:
3758 /* Nothing to do. */
3761 case OFPTYPE_FLOW_MONITOR_CANCEL:
3762 return handle_flow_monitor_cancel(ofconn, oh);
3764 case OFPTYPE_SET_ASYNC_CONFIG:
3765 return handle_nxt_set_async_config(ofconn, oh);
3767 /* Statistics requests. */
3768 case OFPTYPE_DESC_STATS_REQUEST:
3769 return handle_desc_stats_request(ofconn, oh);
3771 case OFPTYPE_FLOW_STATS_REQUEST:
3772 return handle_flow_stats_request(ofconn, oh);
3774 case OFPTYPE_AGGREGATE_STATS_REQUEST:
3775 return handle_aggregate_stats_request(ofconn, oh);
3777 case OFPTYPE_TABLE_STATS_REQUEST:
3778 return handle_table_stats_request(ofconn, oh);
3780 case OFPTYPE_PORT_STATS_REQUEST:
3781 return handle_port_stats_request(ofconn, oh);
3783 case OFPTYPE_QUEUE_STATS_REQUEST:
3784 return handle_queue_stats_request(ofconn, oh);
3786 case OFPTYPE_PORT_DESC_STATS_REQUEST:
3787 return handle_port_desc_stats_request(ofconn, oh);
3789 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
3790 return handle_flow_monitor_request(ofconn, oh);
3794 case OFPTYPE_FEATURES_REPLY:
3795 case OFPTYPE_GET_CONFIG_REPLY:
3796 case OFPTYPE_PACKET_IN:
3797 case OFPTYPE_FLOW_REMOVED:
3798 case OFPTYPE_PORT_STATUS:
3799 case OFPTYPE_BARRIER_REPLY:
3800 case OFPTYPE_DESC_STATS_REPLY:
3801 case OFPTYPE_FLOW_STATS_REPLY:
3802 case OFPTYPE_QUEUE_STATS_REPLY:
3803 case OFPTYPE_PORT_STATS_REPLY:
3804 case OFPTYPE_TABLE_STATS_REPLY:
3805 case OFPTYPE_AGGREGATE_STATS_REPLY:
3806 case OFPTYPE_PORT_DESC_STATS_REPLY:
3807 case OFPTYPE_ROLE_REPLY:
3808 case OFPTYPE_FLOW_MONITOR_PAUSED:
3809 case OFPTYPE_FLOW_MONITOR_RESUMED:
3810 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
3812 return OFPERR_OFPBRC_BAD_TYPE;
3817 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3819 int error = handle_openflow__(ofconn, ofp_msg);
3820 if (error && error != OFPROTO_POSTPONE) {
3821 ofconn_send_error(ofconn, ofp_msg->data, error);
3823 COVERAGE_INC(ofproto_recv_openflow);
3824 return error != OFPROTO_POSTPONE;
3827 /* Asynchronous operations. */
3829 /* Creates and returns a new ofopgroup that is not associated with any
3830 * OpenFlow connection.
3832 * The caller should add operations to the returned group with
3833 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3834 static struct ofopgroup *
3835 ofopgroup_create_unattached(struct ofproto *ofproto)
3837 struct ofopgroup *group = xzalloc(sizeof *group);
3838 group->ofproto = ofproto;
3839 list_init(&group->ofproto_node);
3840 list_init(&group->ops);
3841 list_init(&group->ofconn_node);
3845 /* Creates and returns a new ofopgroup for 'ofproto'.
3847 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3848 * connection. The 'request' and 'buffer_id' arguments are ignored.
3850 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3851 * If the ofopgroup eventually fails, then the error reply will include
3852 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3853 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3855 * The caller should add operations to the returned group with
3856 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3857 static struct ofopgroup *
3858 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3859 const struct ofp_header *request, uint32_t buffer_id)
3861 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3863 size_t request_len = ntohs(request->length);
3865 assert(ofconn_get_ofproto(ofconn) == ofproto);
3867 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3868 group->ofconn = ofconn;
3869 group->request = xmemdup(request, MIN(request_len, 64));
3870 group->buffer_id = buffer_id;
3875 /* Submits 'group' for processing.
3877 * If 'group' contains no operations (e.g. none were ever added, or all of the
3878 * ones that were added completed synchronously), then it is destroyed
3879 * immediately. Otherwise it is added to the ofproto's list of pending
3882 ofopgroup_submit(struct ofopgroup *group)
3884 if (!group->n_running) {
3885 ofopgroup_complete(group);
3887 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3888 group->ofproto->n_pending++;
3893 ofopgroup_complete(struct ofopgroup *group)
3895 struct ofproto *ofproto = group->ofproto;
3897 struct ofconn *abbrev_ofconn;
3898 ovs_be32 abbrev_xid;
3900 struct ofoperation *op, *next_op;
3903 assert(!group->n_running);
3906 LIST_FOR_EACH (op, group_node, &group->ops) {
3913 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
3914 LIST_FOR_EACH (op, group_node, &group->ops) {
3915 if (op->type != OFOPERATION_DELETE) {
3916 struct ofpbuf *packet;
3919 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3923 error = rule_execute(op->rule, in_port, packet);
3930 if (!error && !list_is_empty(&group->ofconn_node)) {
3931 abbrev_ofconn = group->ofconn;
3932 abbrev_xid = group->request->xid;
3934 abbrev_ofconn = NULL;
3935 abbrev_xid = htonl(0);
3937 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
3938 struct rule *rule = op->rule;
3940 if (!op->error && !ofproto_rule_is_hidden(rule)) {
3941 /* Check that we can just cast from ofoperation_type to
3942 * nx_flow_update_event. */
3943 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_ADD
3945 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_DELETE
3947 BUILD_ASSERT_DECL((enum nx_flow_update_event) OFOPERATION_MODIFY
3950 ofmonitor_report(ofproto->connmgr, rule,
3951 (enum nx_flow_update_event) op->type,
3952 op->reason, abbrev_ofconn, abbrev_xid);
3955 rule->pending = NULL;
3958 case OFOPERATION_ADD:
3960 ofproto_rule_destroy__(op->victim);
3961 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3962 == htons(VLAN_VID_MASK)) {
3963 if (ofproto->vlan_bitmap) {
3964 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3966 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3967 bitmap_set1(ofproto->vlan_bitmap, vid);
3968 ofproto->vlans_changed = true;
3971 ofproto->vlans_changed = true;
3975 oftable_substitute_rule(rule, op->victim);
3976 ofproto_rule_destroy__(rule);
3980 case OFOPERATION_DELETE:
3982 ofproto_rule_destroy__(rule);
3986 case OFOPERATION_MODIFY:
3988 rule->modified = time_msec();
3990 rule->flow_cookie = op->flow_cookie;
3992 free(rule->ofpacts);
3993 rule->ofpacts = op->ofpacts;
3994 rule->ofpacts_len = op->ofpacts_len;
3996 op->ofpacts_len = 0;
4005 ofoperation_destroy(op);
4008 ofmonitor_flush(ofproto->connmgr);
4010 if (!list_is_empty(&group->ofproto_node)) {
4011 assert(ofproto->n_pending > 0);
4012 ofproto->n_pending--;
4013 list_remove(&group->ofproto_node);
4015 if (!list_is_empty(&group->ofconn_node)) {
4016 list_remove(&group->ofconn_node);
4018 ofconn_send_error(group->ofconn, group->request, error);
4020 connmgr_retry(ofproto->connmgr);
4022 free(group->request);
4026 /* Initiates a new operation on 'rule', of the specified 'type', within
4027 * 'group'. Prior to calling, 'rule' must not have any pending operation.
4029 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
4030 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
4032 * Returns the newly created ofoperation (which is also available as
4033 * rule->pending). */
4034 static struct ofoperation *
4035 ofoperation_create(struct ofopgroup *group, struct rule *rule,
4036 enum ofoperation_type type,
4037 enum ofp_flow_removed_reason reason)
4039 struct ofproto *ofproto = group->ofproto;
4040 struct ofoperation *op;
4042 assert(!rule->pending);
4044 op = rule->pending = xzalloc(sizeof *op);
4046 list_push_back(&group->ops, &op->group_node);
4049 op->reason = reason;
4050 op->flow_cookie = rule->flow_cookie;
4054 if (type == OFOPERATION_DELETE) {
4055 hmap_insert(&ofproto->deletions, &op->hmap_node,
4056 cls_rule_hash(&rule->cr, rule->table_id));
4063 ofoperation_destroy(struct ofoperation *op)
4065 struct ofopgroup *group = op->group;
4068 op->rule->pending = NULL;
4070 if (op->type == OFOPERATION_DELETE) {
4071 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
4073 list_remove(&op->group_node);
4078 /* Indicates that 'op' completed with status 'error', which is either 0 to
4079 * indicate success or an OpenFlow error code on failure.
4081 * If 'error' is 0, indicating success, the operation will be committed
4082 * permanently to the flow table. There is one interesting subcase:
4084 * - If 'op' is an "add flow" operation that is replacing an existing rule in
4085 * the flow table (the "victim" rule) by a new one, then the caller must
4086 * have uninitialized any derived state in the victim rule, as in step 5 in
4087 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
4088 * performs steps 6 and 7 for the victim rule, most notably by calling its
4089 * ->rule_dealloc() function.
4091 * If 'error' is nonzero, then generally the operation will be rolled back:
4093 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
4094 * restores the original rule. The caller must have uninitialized any
4095 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
4096 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
4097 * and 7 for the new rule, calling its ->rule_dealloc() function.
4099 * - If 'op' is a "modify flow" operation, ofproto restores the original
4102 * - 'op' must not be a "delete flow" operation. Removing a rule is not
4103 * allowed to fail. It must always succeed.
4105 * Please see the large comment in ofproto/ofproto-provider.h titled
4106 * "Asynchronous Operation Support" for more information. */
4108 ofoperation_complete(struct ofoperation *op, enum ofperr error)
4110 struct ofopgroup *group = op->group;
4112 assert(op->rule->pending == op);
4113 assert(group->n_running > 0);
4114 assert(!error || op->type != OFOPERATION_DELETE);
4117 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
4118 ofopgroup_complete(group);
4123 ofoperation_get_victim(struct ofoperation *op)
4125 assert(op->type == OFOPERATION_ADD);
4130 pick_datapath_id(const struct ofproto *ofproto)
4132 const struct ofport *port;
4134 port = ofproto_get_port(ofproto, OFPP_LOCAL);
4136 uint8_t ea[ETH_ADDR_LEN];
4139 error = netdev_get_etheraddr(port->netdev, ea);
4141 return eth_addr_to_uint64(ea);
4143 VLOG_WARN("%s: could not get MAC address for %s (%s)",
4144 ofproto->name, netdev_get_name(port->netdev),
4147 return ofproto->fallback_dpid;
4151 pick_fallback_dpid(void)
4153 uint8_t ea[ETH_ADDR_LEN];
4154 eth_addr_nicira_random(ea);
4155 return eth_addr_to_uint64(ea);
4158 /* Table overflow policy. */
4160 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
4161 * is not configured to evict rules or if the table contains no evictable
4162 * rules. (Rules with 'evictable' set to false or with no timeouts are not
4164 static struct rule *
4165 choose_rule_to_evict(struct oftable *table)
4167 struct eviction_group *evg;
4169 if (!table->eviction_fields) {
4173 /* In the common case, the outer and inner loops here will each be entered
4176 * - The inner loop normally "return"s in its first iteration. If the
4177 * eviction group has any evictable rules, then it always returns in
4180 * - The outer loop only iterates more than once if the largest eviction
4181 * group has no evictable rules.
4183 * - The outer loop can exit only if table's 'max_flows' is all filled up
4184 * by unevictable rules'. */
4185 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
4188 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
4189 if (rule->evictable) {
4198 /* Searches 'ofproto' for tables that have more flows than their configured
4199 * maximum and that have flow eviction enabled, and evicts as many flows as
4200 * necessary and currently feasible from them.
4202 * This triggers only when an OpenFlow table has N flows in it and then the
4203 * client configures a maximum number of flows less than N. */
4205 ofproto_evict(struct ofproto *ofproto)
4207 struct ofopgroup *group;
4208 struct oftable *table;
4210 group = ofopgroup_create_unattached(ofproto);
4211 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
4212 while (classifier_count(&table->cls) > table->max_flows
4213 && table->eviction_fields) {
4216 rule = choose_rule_to_evict(table);
4217 if (!rule || rule->pending) {
4221 ofoperation_create(group, rule,
4222 OFOPERATION_DELETE, OFPRR_EVICTION);
4223 oftable_remove_rule(rule);
4224 ofproto->ofproto_class->rule_destruct(rule);
4227 ofopgroup_submit(group);
4230 /* Eviction groups. */
4232 /* Returns the priority to use for an eviction_group that contains 'n_rules'
4233 * rules. The priority contains low-order random bits to ensure that eviction
4234 * groups with the same number of rules are prioritized randomly. */
4236 eviction_group_priority(size_t n_rules)
4238 uint16_t size = MIN(UINT16_MAX, n_rules);
4239 return (size << 16) | random_uint16();
4242 /* Updates 'evg', an eviction_group within 'table', following a change that
4243 * adds or removes rules in 'evg'. */
4245 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
4247 heap_change(&table->eviction_groups_by_size, &evg->size_node,
4248 eviction_group_priority(heap_count(&evg->rules)));
4251 /* Destroys 'evg', an eviction_group within 'table':
4253 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
4254 * rules themselves, just removes them from the eviction group.)
4256 * - Removes 'evg' from 'table'.
4260 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
4262 while (!heap_is_empty(&evg->rules)) {
4265 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
4266 rule->eviction_group = NULL;
4268 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
4269 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
4270 heap_destroy(&evg->rules);
4274 /* Removes 'rule' from its eviction group, if any. */
4276 eviction_group_remove_rule(struct rule *rule)
4278 if (rule->eviction_group) {
4279 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4280 struct eviction_group *evg = rule->eviction_group;
4282 rule->eviction_group = NULL;
4283 heap_remove(&evg->rules, &rule->evg_node);
4284 if (heap_is_empty(&evg->rules)) {
4285 eviction_group_destroy(table, evg);
4287 eviction_group_resized(table, evg);
4292 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
4293 * returns the hash value. */
4295 eviction_group_hash_rule(struct rule *rule)
4297 struct oftable *table = &rule->ofproto->tables[rule->table_id];
4298 const struct mf_subfield *sf;
4301 hash = table->eviction_group_id_basis;
4302 for (sf = table->eviction_fields;
4303 sf < &table->eviction_fields[table->n_eviction_fields];
4306 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
4307 union mf_value value;
4309 mf_get_value(sf->field, &rule->cr.flow, &value);
4311 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
4313 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
4314 unsigned int start = sf->ofs + sf->n_bits;
4315 bitwise_zero(&value, sf->field->n_bytes, start,
4316 sf->field->n_bytes * 8 - start);
4318 hash = hash_bytes(&value, sf->field->n_bytes, hash);
4320 hash = hash_int(hash, 0);
4327 /* Returns an eviction group within 'table' with the given 'id', creating one
4329 static struct eviction_group *
4330 eviction_group_find(struct oftable *table, uint32_t id)
4332 struct eviction_group *evg;
4334 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
4338 evg = xmalloc(sizeof *evg);
4339 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
4340 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
4341 eviction_group_priority(0));
4342 heap_init(&evg->rules);
4347 /* Returns an eviction priority for 'rule'. The return value should be
4348 * interpreted so that higher priorities make a rule more attractive candidates
4351 rule_eviction_priority(struct rule *rule)
4353 long long int hard_expiration;
4354 long long int idle_expiration;
4355 long long int expiration;
4356 uint32_t expiration_offset;
4358 /* Calculate time of expiration. */
4359 hard_expiration = (rule->hard_timeout
4360 ? rule->modified + rule->hard_timeout * 1000
4362 idle_expiration = (rule->idle_timeout
4363 ? rule->used + rule->idle_timeout * 1000
4365 expiration = MIN(hard_expiration, idle_expiration);
4366 if (expiration == LLONG_MAX) {
4370 /* Calculate the time of expiration as a number of (approximate) seconds
4371 * after program startup.
4373 * This should work OK for program runs that last UINT32_MAX seconds or
4374 * less. Therefore, please restart OVS at least once every 136 years. */
4375 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
4377 /* Invert the expiration offset because we're using a max-heap. */
4378 return UINT32_MAX - expiration_offset;
4381 /* Adds 'rule' to an appropriate eviction group for its oftable's
4382 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4383 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4386 * The caller must ensure that 'rule' is not already in an eviction group. */
4388 eviction_group_add_rule(struct rule *rule)
4390 struct ofproto *ofproto = rule->ofproto;
4391 struct oftable *table = &ofproto->tables[rule->table_id];
4393 if (table->eviction_fields
4394 && (rule->hard_timeout || rule->idle_timeout)) {
4395 struct eviction_group *evg;
4397 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4399 rule->eviction_group = evg;
4400 heap_insert(&evg->rules, &rule->evg_node,
4401 rule_eviction_priority(rule));
4402 eviction_group_resized(table, evg);
4408 /* Initializes 'table'. */
4410 oftable_init(struct oftable *table)
4412 memset(table, 0, sizeof *table);
4413 classifier_init(&table->cls);
4414 table->max_flows = UINT_MAX;
4417 /* Destroys 'table', including its classifier and eviction groups.
4419 * The caller is responsible for freeing 'table' itself. */
4421 oftable_destroy(struct oftable *table)
4423 assert(classifier_is_empty(&table->cls));
4424 oftable_disable_eviction(table);
4425 classifier_destroy(&table->cls);
4429 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4430 * string, then 'table' will use its default name.
4432 * This only affects the name exposed for a table exposed through the OpenFlow
4433 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4435 oftable_set_name(struct oftable *table, const char *name)
4437 if (name && name[0]) {
4438 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4439 if (!table->name || strncmp(name, table->name, len)) {
4441 table->name = xmemdup0(name, len);
4449 /* oftables support a choice of two policies when adding a rule would cause the
4450 * number of flows in the table to exceed the configured maximum number: either
4451 * they can refuse to add the new flow or they can evict some existing flow.
4452 * This function configures the former policy on 'table'. */
4454 oftable_disable_eviction(struct oftable *table)
4456 if (table->eviction_fields) {
4457 struct eviction_group *evg, *next;
4459 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4460 &table->eviction_groups_by_id) {
4461 eviction_group_destroy(table, evg);
4463 hmap_destroy(&table->eviction_groups_by_id);
4464 heap_destroy(&table->eviction_groups_by_size);
4466 free(table->eviction_fields);
4467 table->eviction_fields = NULL;
4468 table->n_eviction_fields = 0;
4472 /* oftables support a choice of two policies when adding a rule would cause the
4473 * number of flows in the table to exceed the configured maximum number: either
4474 * they can refuse to add the new flow or they can evict some existing flow.
4475 * This function configures the latter policy on 'table', with fairness based
4476 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4477 * 'n_fields' as 0 disables fairness.) */
4479 oftable_enable_eviction(struct oftable *table,
4480 const struct mf_subfield *fields, size_t n_fields)
4482 struct cls_cursor cursor;
4485 if (table->eviction_fields
4486 && n_fields == table->n_eviction_fields
4488 || !memcmp(fields, table->eviction_fields,
4489 n_fields * sizeof *fields))) {
4494 oftable_disable_eviction(table);
4496 table->n_eviction_fields = n_fields;
4497 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4499 table->eviction_group_id_basis = random_uint32();
4500 hmap_init(&table->eviction_groups_by_id);
4501 heap_init(&table->eviction_groups_by_size);
4503 cls_cursor_init(&cursor, &table->cls, NULL);
4504 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4505 eviction_group_add_rule(rule);
4509 /* Removes 'rule' from the oftable that contains it. */
4511 oftable_remove_rule(struct rule *rule)
4513 struct ofproto *ofproto = rule->ofproto;
4514 struct oftable *table = &ofproto->tables[rule->table_id];
4516 classifier_remove(&table->cls, &rule->cr);
4517 eviction_group_remove_rule(rule);
4520 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4521 * oftable that has an identical cls_rule. Returns the rule that was removed,
4522 * if any, and otherwise NULL. */
4523 static struct rule *
4524 oftable_replace_rule(struct rule *rule)
4526 struct ofproto *ofproto = rule->ofproto;
4527 struct oftable *table = &ofproto->tables[rule->table_id];
4528 struct rule *victim;
4530 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4532 eviction_group_remove_rule(victim);
4534 eviction_group_add_rule(rule);
4538 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4540 oftable_substitute_rule(struct rule *old, struct rule *new)
4543 oftable_replace_rule(new);
4545 oftable_remove_rule(old);
4549 /* unixctl commands. */
4552 ofproto_lookup(const char *name)
4554 struct ofproto *ofproto;
4556 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4558 if (!strcmp(ofproto->name, name)) {
4566 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4567 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4569 struct ofproto *ofproto;
4573 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4574 ds_put_format(&results, "%s\n", ofproto->name);
4576 unixctl_command_reply(conn, ds_cstr(&results));
4577 ds_destroy(&results);
4581 ofproto_unixctl_init(void)
4583 static bool registered;
4589 unixctl_command_register("ofproto/list", "", 0, 0,
4590 ofproto_unixctl_list, NULL);
4593 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4595 * This is deprecated. It is only for compatibility with broken device drivers
4596 * in old versions of Linux that do not properly support VLANs when VLAN
4597 * devices are not used. When broken device drivers are no longer in
4598 * widespread use, we will delete these interfaces. */
4600 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4601 * (exactly) by an OpenFlow rule in 'ofproto'. */
4603 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4605 const struct oftable *oftable;
4607 free(ofproto->vlan_bitmap);
4608 ofproto->vlan_bitmap = bitmap_allocate(4096);
4609 ofproto->vlans_changed = false;
4611 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4612 const struct cls_table *table;
4614 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4615 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4616 == htons(VLAN_VID_MASK)) {
4617 const struct cls_rule *rule;
4619 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4620 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4621 bitmap_set1(vlan_bitmap, vid);
4622 bitmap_set1(ofproto->vlan_bitmap, vid);
4629 /* Returns true if new VLANs have come into use by the flow table since the
4630 * last call to ofproto_get_vlan_usage().
4632 * We don't track when old VLANs stop being used. */
4634 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4636 return ofproto->vlans_changed;
4639 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4640 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4641 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4642 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4643 * then the VLAN device is un-enslaved. */
4645 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4646 uint16_t realdev_ofp_port, int vid)
4648 struct ofport *ofport;
4651 assert(vlandev_ofp_port != realdev_ofp_port);
4653 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4655 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4656 ofproto->name, vlandev_ofp_port);
4660 if (!ofproto->ofproto_class->set_realdev) {
4661 if (!vlandev_ofp_port) {
4664 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4668 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4670 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4671 ofproto->name, vlandev_ofp_port,
4672 netdev_get_name(ofport->netdev), strerror(error));