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"
37 #include "ofp-print.h"
40 #include "ofproto-provider.h"
41 #include "openflow/nicira-ext.h"
42 #include "openflow/openflow.h"
46 #include "poll-loop.h"
52 #include "unaligned.h"
56 VLOG_DEFINE_THIS_MODULE(ofproto);
58 COVERAGE_DEFINE(ofproto_error);
59 COVERAGE_DEFINE(ofproto_flush);
60 COVERAGE_DEFINE(ofproto_no_packet_in);
61 COVERAGE_DEFINE(ofproto_packet_out);
62 COVERAGE_DEFINE(ofproto_queue_req);
63 COVERAGE_DEFINE(ofproto_recv_openflow);
64 COVERAGE_DEFINE(ofproto_reinit_ports);
65 COVERAGE_DEFINE(ofproto_uninstallable);
66 COVERAGE_DEFINE(ofproto_update_port);
69 S_OPENFLOW, /* Processing OpenFlow commands. */
70 S_EVICT, /* Evicting flows from over-limit tables. */
71 S_FLUSH, /* Deleting all flow table rules. */
74 enum ofoperation_type {
80 /* A single OpenFlow request can execute any number of operations. The
81 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
82 * ofconn to which an error reply should be sent if necessary.
84 * ofproto initiates some operations internally. These operations are still
85 * assigned to groups but will not have an associated ofconn. */
87 struct ofproto *ofproto; /* Owning ofproto. */
88 struct list ofproto_node; /* In ofproto's "pending" list. */
89 struct list ops; /* List of "struct ofoperation"s. */
91 /* Data needed to send OpenFlow reply on failure or to send a buffered
94 * If list_is_empty(ofconn_node) then this ofopgroup never had an
95 * associated ofconn or its ofconn's connection dropped after it initiated
96 * the operation. In the latter case 'ofconn' is a wild pointer that
97 * refers to freed memory, so the 'ofconn' member must be used only if
98 * !list_is_empty(ofconn_node).
100 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
101 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
102 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
103 uint32_t buffer_id; /* Buffer id from original request. */
104 int error; /* 0 if no error yet, otherwise error code. */
107 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
108 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
109 const struct ofp_header *,
111 static void ofopgroup_submit(struct ofopgroup *);
112 static void ofopgroup_destroy(struct ofopgroup *);
114 /* A single flow table operation. */
116 struct ofopgroup *group; /* Owning group. */
117 struct list group_node; /* In ofopgroup's "ops" list. */
118 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
119 struct rule *rule; /* Rule being operated upon. */
120 enum ofoperation_type type; /* Type of operation. */
121 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
122 struct ofpact *ofpacts; /* OFOPERATION_MODIFYING: Replaced actions. */
123 size_t ofpacts_len; /* OFOPERATION_MODIFYING: Bytes of ofpacts. */
124 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
127 static void ofoperation_create(struct ofopgroup *, struct rule *,
128 enum ofoperation_type);
129 static void ofoperation_destroy(struct ofoperation *);
132 static void oftable_init(struct oftable *);
133 static void oftable_destroy(struct oftable *);
135 static void oftable_set_name(struct oftable *, const char *name);
137 static void oftable_disable_eviction(struct oftable *);
138 static void oftable_enable_eviction(struct oftable *,
139 const struct mf_subfield *fields,
142 static void oftable_remove_rule(struct rule *);
143 static struct rule *oftable_replace_rule(struct rule *);
144 static void oftable_substitute_rule(struct rule *old, struct rule *new);
146 /* A set of rules within a single OpenFlow table (oftable) that have the same
147 * values for the oftable's eviction_fields. A rule to be evicted, when one is
148 * needed, is taken from the eviction group that contains the greatest number
151 * An oftable owns any number of eviction groups, each of which contains any
154 * Membership in an eviction group is imprecise, based on the hash of the
155 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
156 * That is, if two rules have different eviction_fields, but those
157 * eviction_fields hash to the same value, then they will belong to the same
158 * eviction_group anyway.
160 * (When eviction is not enabled on an oftable, we don't track any eviction
161 * groups, to save time and space.) */
162 struct eviction_group {
163 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
164 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
165 struct heap rules; /* Contains "struct rule"s. */
168 static struct rule *choose_rule_to_evict(struct oftable *);
169 static void ofproto_evict(struct ofproto *);
170 static uint32_t rule_eviction_priority(struct rule *);
173 static void ofport_destroy__(struct ofport *);
174 static void ofport_destroy(struct ofport *);
176 static void update_port(struct ofproto *, const char *devname);
177 static int init_ports(struct ofproto *);
178 static void reinit_ports(struct ofproto *);
181 static void ofproto_rule_destroy__(struct rule *);
182 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
183 static bool rule_is_modifiable(const struct rule *);
184 static bool rule_is_hidden(const struct rule *);
187 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
188 const struct ofputil_flow_mod *,
189 const struct ofp_header *);
190 static void delete_flow__(struct rule *, struct ofopgroup *);
191 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
192 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
193 const struct ofputil_flow_mod *,
194 const struct ofp_header *);
197 static uint64_t pick_datapath_id(const struct ofproto *);
198 static uint64_t pick_fallback_dpid(void);
199 static void ofproto_destroy__(struct ofproto *);
200 static void update_mtu(struct ofproto *, struct ofport *);
203 static void ofproto_unixctl_init(void);
205 /* All registered ofproto classes, in probe order. */
206 static const struct ofproto_class **ofproto_classes;
207 static size_t n_ofproto_classes;
208 static size_t allocated_ofproto_classes;
210 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
211 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
216 ofproto_initialize(void)
222 ofproto_class_register(&ofproto_dpif_class);
226 /* 'type' should be a normalized datapath type, as returned by
227 * ofproto_normalize_type(). Returns the corresponding ofproto_class
228 * structure, or a null pointer if there is none registered for 'type'. */
229 static const struct ofproto_class *
230 ofproto_class_find__(const char *type)
234 ofproto_initialize();
235 for (i = 0; i < n_ofproto_classes; i++) {
236 const struct ofproto_class *class = ofproto_classes[i];
241 class->enumerate_types(&types);
242 found = sset_contains(&types, type);
243 sset_destroy(&types);
249 VLOG_WARN("unknown datapath type %s", type);
253 /* Registers a new ofproto class. After successful registration, new ofprotos
254 * of that type can be created using ofproto_create(). */
256 ofproto_class_register(const struct ofproto_class *new_class)
260 for (i = 0; i < n_ofproto_classes; i++) {
261 if (ofproto_classes[i] == new_class) {
266 if (n_ofproto_classes >= allocated_ofproto_classes) {
267 ofproto_classes = x2nrealloc(ofproto_classes,
268 &allocated_ofproto_classes,
269 sizeof *ofproto_classes);
271 ofproto_classes[n_ofproto_classes++] = new_class;
275 /* Unregisters a datapath provider. 'type' must have been previously
276 * registered and not currently be in use by any ofprotos. After
277 * unregistration new datapaths of that type cannot be opened using
278 * ofproto_create(). */
280 ofproto_class_unregister(const struct ofproto_class *class)
284 for (i = 0; i < n_ofproto_classes; i++) {
285 if (ofproto_classes[i] == class) {
286 for (i++; i < n_ofproto_classes; i++) {
287 ofproto_classes[i - 1] = ofproto_classes[i];
293 VLOG_WARN("attempted to unregister an ofproto class that is not "
298 /* Clears 'types' and enumerates all registered ofproto types into it. The
299 * caller must first initialize the sset. */
301 ofproto_enumerate_types(struct sset *types)
305 ofproto_initialize();
306 for (i = 0; i < n_ofproto_classes; i++) {
307 ofproto_classes[i]->enumerate_types(types);
311 /* Returns the fully spelled out name for the given ofproto 'type'.
313 * Normalized type string can be compared with strcmp(). Unnormalized type
314 * string might be the same even if they have different spellings. */
316 ofproto_normalize_type(const char *type)
318 return type && type[0] ? type : "system";
321 /* Clears 'names' and enumerates the names of all known created ofprotos with
322 * the given 'type'. The caller must first initialize the sset. Returns 0 if
323 * successful, otherwise a positive errno value.
325 * Some kinds of datapaths might not be practically enumerable. This is not
326 * considered an error. */
328 ofproto_enumerate_names(const char *type, struct sset *names)
330 const struct ofproto_class *class = ofproto_class_find__(type);
331 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
335 ofproto_create(const char *datapath_name, const char *datapath_type,
336 struct ofproto **ofprotop)
338 const struct ofproto_class *class;
339 struct ofproto *ofproto;
344 ofproto_initialize();
345 ofproto_unixctl_init();
347 datapath_type = ofproto_normalize_type(datapath_type);
348 class = ofproto_class_find__(datapath_type);
350 VLOG_WARN("could not create datapath %s of unknown type %s",
351 datapath_name, datapath_type);
355 ofproto = class->alloc();
357 VLOG_ERR("failed to allocate datapath %s of type %s",
358 datapath_name, datapath_type);
363 memset(ofproto, 0, sizeof *ofproto);
364 ofproto->ofproto_class = class;
365 ofproto->name = xstrdup(datapath_name);
366 ofproto->type = xstrdup(datapath_type);
367 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
368 hash_string(ofproto->name, 0));
369 ofproto->datapath_id = 0;
370 ofproto_set_flow_eviction_threshold(ofproto,
371 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
372 ofproto->forward_bpdu = false;
373 ofproto->fallback_dpid = pick_fallback_dpid();
374 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
375 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
376 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
377 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
378 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
379 ofproto->frag_handling = OFPC_FRAG_NORMAL;
380 hmap_init(&ofproto->ports);
381 shash_init(&ofproto->port_by_name);
382 ofproto->tables = NULL;
383 ofproto->n_tables = 0;
384 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
385 ofproto->state = S_OPENFLOW;
386 list_init(&ofproto->pending);
387 ofproto->n_pending = 0;
388 hmap_init(&ofproto->deletions);
389 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
390 ofproto->first_op = ofproto->last_op = LLONG_MIN;
391 ofproto->next_op_report = LLONG_MAX;
392 ofproto->op_backoff = LLONG_MIN;
393 ofproto->vlan_bitmap = NULL;
394 ofproto->vlans_changed = false;
395 ofproto->min_mtu = INT_MAX;
397 error = ofproto->ofproto_class->construct(ofproto);
399 VLOG_ERR("failed to open datapath %s: %s",
400 datapath_name, strerror(error));
401 ofproto_destroy__(ofproto);
405 assert(ofproto->n_tables);
407 ofproto->datapath_id = pick_datapath_id(ofproto);
408 VLOG_INFO("%s: using datapath ID %016"PRIx64,
409 ofproto->name, ofproto->datapath_id);
417 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
419 struct oftable *table;
421 assert(!ofproto->n_tables);
422 assert(n_tables >= 1 && n_tables <= 255);
424 ofproto->n_tables = n_tables;
425 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
426 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
432 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
434 uint64_t old_dpid = p->datapath_id;
435 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
436 if (p->datapath_id != old_dpid) {
437 VLOG_INFO("%s: datapath ID changed to %016"PRIx64,
438 p->name, p->datapath_id);
440 /* Force all active connections to reconnect, since there is no way to
441 * notify a controller that the datapath ID has changed. */
442 ofproto_reconnect_controllers(p);
447 ofproto_set_controllers(struct ofproto *p,
448 const struct ofproto_controller *controllers,
449 size_t n_controllers)
451 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
455 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
457 connmgr_set_fail_mode(p->connmgr, fail_mode);
460 /* Drops the connections between 'ofproto' and all of its controllers, forcing
461 * them to reconnect. */
463 ofproto_reconnect_controllers(struct ofproto *ofproto)
465 connmgr_reconnect(ofproto->connmgr);
468 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
469 * in-band control should guarantee access, in the same way that in-band
470 * control guarantees access to OpenFlow controllers. */
472 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
473 const struct sockaddr_in *extras, size_t n)
475 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
478 /* Sets the OpenFlow queue used by flows set up by in-band control on
479 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
480 * flows will use the default queue. */
482 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
484 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
487 /* Sets the number of flows at which eviction from the kernel flow table
490 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
492 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
493 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
495 ofproto->flow_eviction_threshold = threshold;
499 /* If forward_bpdu is true, the NORMAL action will forward frames with
500 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
501 * the NORMAL action will drop these frames. */
503 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
505 bool old_val = ofproto->forward_bpdu;
506 ofproto->forward_bpdu = forward_bpdu;
507 if (old_val != ofproto->forward_bpdu) {
508 if (ofproto->ofproto_class->forward_bpdu_changed) {
509 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
514 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
515 * 'idle_time', in seconds. */
517 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
519 if (ofproto->ofproto_class->set_mac_idle_time) {
520 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
525 ofproto_set_desc(struct ofproto *p,
526 const char *mfr_desc, const char *hw_desc,
527 const char *sw_desc, const char *serial_desc,
530 struct ofp_desc_stats *ods;
533 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
534 VLOG_WARN("%s: truncating mfr_desc, must be less than %zu bytes",
535 p->name, sizeof ods->mfr_desc);
538 p->mfr_desc = xstrdup(mfr_desc);
541 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
542 VLOG_WARN("%s: truncating hw_desc, must be less than %zu bytes",
543 p->name, sizeof ods->hw_desc);
546 p->hw_desc = xstrdup(hw_desc);
549 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
550 VLOG_WARN("%s: truncating sw_desc, must be less than %zu bytes",
551 p->name, sizeof ods->sw_desc);
554 p->sw_desc = xstrdup(sw_desc);
557 if (strlen(serial_desc) >= sizeof ods->serial_num) {
558 VLOG_WARN("%s: truncating serial_desc, must be less than %zu "
559 "bytes", p->name, sizeof ods->serial_num);
561 free(p->serial_desc);
562 p->serial_desc = xstrdup(serial_desc);
565 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
566 VLOG_WARN("%s: truncating dp_desc, must be less than %zu bytes",
567 p->name, sizeof ods->dp_desc);
570 p->dp_desc = xstrdup(dp_desc);
575 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
577 return connmgr_set_snoops(ofproto->connmgr, snoops);
581 ofproto_set_netflow(struct ofproto *ofproto,
582 const struct netflow_options *nf_options)
584 if (nf_options && sset_is_empty(&nf_options->collectors)) {
588 if (ofproto->ofproto_class->set_netflow) {
589 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
591 return nf_options ? EOPNOTSUPP : 0;
596 ofproto_set_sflow(struct ofproto *ofproto,
597 const struct ofproto_sflow_options *oso)
599 if (oso && sset_is_empty(&oso->targets)) {
603 if (ofproto->ofproto_class->set_sflow) {
604 return ofproto->ofproto_class->set_sflow(ofproto, oso);
606 return oso ? EOPNOTSUPP : 0;
610 /* Spanning Tree Protocol (STP) configuration. */
612 /* Configures STP on 'ofproto' using the settings defined in 's'. If
613 * 's' is NULL, disables STP.
615 * Returns 0 if successful, otherwise a positive errno value. */
617 ofproto_set_stp(struct ofproto *ofproto,
618 const struct ofproto_stp_settings *s)
620 return (ofproto->ofproto_class->set_stp
621 ? ofproto->ofproto_class->set_stp(ofproto, s)
625 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
626 * 'enabled' member of 's' is false, then the other members are not
629 * Returns 0 if successful, otherwise a positive errno value. */
631 ofproto_get_stp_status(struct ofproto *ofproto,
632 struct ofproto_stp_status *s)
634 return (ofproto->ofproto_class->get_stp_status
635 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
639 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
640 * in 's'. The caller is responsible for assigning STP port numbers
641 * (using the 'port_num' member in the range of 1 through 255, inclusive)
642 * and ensuring there are no duplicates. If the 's' is NULL, then STP
643 * is disabled on the port.
645 * Returns 0 if successful, otherwise a positive errno value.*/
647 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
648 const struct ofproto_port_stp_settings *s)
650 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
652 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
653 ofproto->name, ofp_port);
657 return (ofproto->ofproto_class->set_stp_port
658 ? ofproto->ofproto_class->set_stp_port(ofport, s)
662 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
663 * 's'. If the 'enabled' member in 's' is false, then the other members
664 * are not meaningful.
666 * Returns 0 if successful, otherwise a positive errno value.*/
668 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
669 struct ofproto_port_stp_status *s)
671 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
673 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
674 "port %"PRIu16, ofproto->name, ofp_port);
678 return (ofproto->ofproto_class->get_stp_port_status
679 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
683 /* Queue DSCP configuration. */
685 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
686 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
687 * to implement QoS. Instead, it is used to implement features which require
688 * knowledge of what queues exist on a port, and some basic information about
691 * Returns 0 if successful, otherwise a positive errno value. */
693 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
694 const struct ofproto_port_queue *queues,
697 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
700 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
701 ofproto->name, ofp_port);
705 return (ofproto->ofproto_class->set_queues
706 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
710 /* Connectivity Fault Management configuration. */
712 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
714 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
716 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
717 if (ofport && ofproto->ofproto_class->set_cfm) {
718 ofproto->ofproto_class->set_cfm(ofport, NULL);
722 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
723 * basic configuration from the configuration members in 'cfm', and the remote
724 * maintenance point ID from remote_mpid. Ignores the statistics members of
727 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
729 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
730 const struct cfm_settings *s)
732 struct ofport *ofport;
735 ofport = ofproto_get_port(ofproto, ofp_port);
737 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
738 ofproto->name, ofp_port);
742 /* XXX: For configuration simplicity, we only support one remote_mpid
743 * outside of the CFM module. It's not clear if this is the correct long
744 * term solution or not. */
745 error = (ofproto->ofproto_class->set_cfm
746 ? ofproto->ofproto_class->set_cfm(ofport, s)
749 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
750 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
755 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
756 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
757 * 0 if LACP partner information is not current (generally indicating a
758 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
760 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
762 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
763 return (ofport && ofproto->ofproto_class->port_is_lacp_current
764 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
770 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
771 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
772 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
773 * configuration plus, if there is more than one slave, a bonding
776 * If 'aux' is already registered then this function updates its configuration
777 * to 's'. Otherwise, this function registers a new bundle.
779 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
782 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
783 const struct ofproto_bundle_settings *s)
785 return (ofproto->ofproto_class->bundle_set
786 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
790 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
791 * If no such bundle has been registered, this has no effect. */
793 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
795 return ofproto_bundle_register(ofproto, aux, NULL);
799 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
800 * If 'aux' is already registered then this function updates its configuration
801 * to 's'. Otherwise, this function registers a new mirror. */
803 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
804 const struct ofproto_mirror_settings *s)
806 return (ofproto->ofproto_class->mirror_set
807 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
811 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
812 * If no mirror has been registered, this has no effect. */
814 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
816 return ofproto_mirror_register(ofproto, aux, NULL);
819 /* Retrieves statistics from mirror associated with client data pointer
820 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
821 * 'bytes', respectively. If a particular counters is not supported,
822 * the appropriate argument is set to UINT64_MAX. */
824 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
825 uint64_t *packets, uint64_t *bytes)
827 if (!ofproto->ofproto_class->mirror_get_stats) {
828 *packets = *bytes = UINT64_MAX;
832 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
836 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
837 * which all packets are flooded, instead of using MAC learning. If
838 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
840 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
843 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
845 return (ofproto->ofproto_class->set_flood_vlans
846 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
850 /* Returns true if 'aux' is a registered bundle that is currently in use as the
851 * output for a mirror. */
853 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
855 return (ofproto->ofproto_class->is_mirror_output_bundle
856 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
860 /* Configuration of OpenFlow tables. */
862 /* Returns the number of OpenFlow tables in 'ofproto'. */
864 ofproto_get_n_tables(const struct ofproto *ofproto)
866 return ofproto->n_tables;
869 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
870 * settings from 's'. 'table_id' must be in the range 0 through the number of
871 * OpenFlow tables in 'ofproto' minus 1, inclusive.
873 * For read-only tables, only the name may be configured. */
875 ofproto_configure_table(struct ofproto *ofproto, int table_id,
876 const struct ofproto_table_settings *s)
878 struct oftable *table;
880 assert(table_id >= 0 && table_id < ofproto->n_tables);
881 table = &ofproto->tables[table_id];
883 oftable_set_name(table, s->name);
885 if (table->flags & OFTABLE_READONLY) {
890 oftable_enable_eviction(table, s->groups, s->n_groups);
892 oftable_disable_eviction(table);
895 table->max_flows = s->max_flows;
896 if (classifier_count(&table->cls) > table->max_flows
897 && table->eviction_fields) {
898 /* 'table' contains more flows than allowed. We might not be able to
899 * evict them right away because of the asynchronous nature of flow
900 * table changes. Schedule eviction for later. */
901 switch (ofproto->state) {
903 ofproto->state = S_EVICT;
907 /* We're already deleting flows, nothing more to do. */
914 ofproto_has_snoops(const struct ofproto *ofproto)
916 return connmgr_has_snoops(ofproto->connmgr);
920 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
922 connmgr_get_snoops(ofproto->connmgr, snoops);
926 ofproto_flush__(struct ofproto *ofproto)
928 struct ofopgroup *group;
929 struct oftable *table;
931 if (ofproto->ofproto_class->flush) {
932 ofproto->ofproto_class->flush(ofproto);
935 group = ofopgroup_create_unattached(ofproto);
936 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
937 struct rule *rule, *next_rule;
938 struct cls_cursor cursor;
940 if (table->flags & OFTABLE_HIDDEN) {
944 cls_cursor_init(&cursor, &table->cls, NULL);
945 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
946 if (!rule->pending) {
947 ofoperation_create(group, rule, OFOPERATION_DELETE);
948 oftable_remove_rule(rule);
949 ofproto->ofproto_class->rule_destruct(rule);
953 ofopgroup_submit(group);
957 ofproto_destroy__(struct ofproto *ofproto)
959 struct oftable *table;
961 assert(list_is_empty(&ofproto->pending));
962 assert(!ofproto->n_pending);
964 connmgr_destroy(ofproto->connmgr);
966 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
969 free(ofproto->mfr_desc);
970 free(ofproto->hw_desc);
971 free(ofproto->sw_desc);
972 free(ofproto->serial_desc);
973 free(ofproto->dp_desc);
974 hmap_destroy(&ofproto->ports);
975 shash_destroy(&ofproto->port_by_name);
977 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
978 oftable_destroy(table);
980 free(ofproto->tables);
982 hmap_destroy(&ofproto->deletions);
984 free(ofproto->vlan_bitmap);
986 ofproto->ofproto_class->dealloc(ofproto);
990 ofproto_destroy(struct ofproto *p)
992 struct ofport *ofport, *next_ofport;
999 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1000 ofport_destroy(ofport);
1003 p->ofproto_class->destruct(p);
1004 ofproto_destroy__(p);
1007 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1008 * kernel datapath, for example, this destroys the datapath in the kernel, and
1009 * with the netdev-based datapath, it tears down the data structures that
1010 * represent the datapath.
1012 * The datapath should not be currently open as an ofproto. */
1014 ofproto_delete(const char *name, const char *type)
1016 const struct ofproto_class *class = ofproto_class_find__(type);
1017 return (!class ? EAFNOSUPPORT
1018 : !class->del ? EACCES
1019 : class->del(type, name));
1023 process_port_change(struct ofproto *ofproto, int error, char *devname)
1025 if (error == ENOBUFS) {
1026 reinit_ports(ofproto);
1027 } else if (!error) {
1028 update_port(ofproto, devname);
1034 ofproto_run(struct ofproto *p)
1036 struct sset changed_netdevs;
1037 const char *changed_netdev;
1038 struct ofport *ofport;
1041 error = p->ofproto_class->run(p);
1042 if (error && error != EAGAIN) {
1043 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
1046 if (p->ofproto_class->port_poll) {
1049 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1050 process_port_change(p, error, devname);
1054 /* Update OpenFlow port status for any port whose netdev has changed.
1056 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1057 * destroyed, so it's not safe to update ports directly from the
1058 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1059 * need this two-phase approach. */
1060 sset_init(&changed_netdevs);
1061 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1062 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1063 if (ofport->change_seq != change_seq) {
1064 ofport->change_seq = change_seq;
1065 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1068 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1069 update_port(p, changed_netdev);
1071 sset_destroy(&changed_netdevs);
1075 connmgr_run(p->connmgr, handle_openflow);
1079 connmgr_run(p->connmgr, NULL);
1081 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1082 p->state = S_OPENFLOW;
1087 connmgr_run(p->connmgr, NULL);
1089 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1090 connmgr_flushed(p->connmgr);
1091 p->state = S_OPENFLOW;
1099 if (time_msec() >= p->next_op_report) {
1100 long long int ago = (time_msec() - p->first_op) / 1000;
1101 long long int interval = (p->last_op - p->first_op) / 1000;
1105 ds_put_format(&s, "%d flow_mods ",
1106 p->n_add + p->n_delete + p->n_modify);
1107 if (interval == ago) {
1108 ds_put_format(&s, "in the last %lld s", ago);
1109 } else if (interval) {
1110 ds_put_format(&s, "in the %lld s starting %lld s ago",
1113 ds_put_format(&s, "%lld s ago", ago);
1116 ds_put_cstr(&s, " (");
1118 ds_put_format(&s, "%d adds, ", p->n_add);
1121 ds_put_format(&s, "%d deletes, ", p->n_delete);
1124 ds_put_format(&s, "%d modifications, ", p->n_modify);
1127 ds_put_char(&s, ')');
1129 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1132 p->n_add = p->n_delete = p->n_modify = 0;
1133 p->next_op_report = LLONG_MAX;
1139 /* Performs periodic activity required by 'ofproto' that needs to be done
1140 * with the least possible latency.
1142 * It makes sense to call this function a couple of times per poll loop, to
1143 * provide a significant performance boost on some benchmarks with the
1144 * ofproto-dpif implementation. */
1146 ofproto_run_fast(struct ofproto *p)
1150 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1151 if (error && error != EAGAIN) {
1152 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1153 p->name, strerror(error));
1159 ofproto_wait(struct ofproto *p)
1161 struct ofport *ofport;
1163 p->ofproto_class->wait(p);
1164 if (p->ofproto_class->port_poll_wait) {
1165 p->ofproto_class->port_poll_wait(p);
1168 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1169 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1170 poll_immediate_wake();
1176 connmgr_wait(p->connmgr, true);
1181 connmgr_wait(p->connmgr, false);
1182 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1183 poll_immediate_wake();
1190 ofproto_is_alive(const struct ofproto *p)
1192 return connmgr_has_controllers(p->connmgr);
1195 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1196 * memory_report(). */
1198 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1200 const struct oftable *table;
1201 unsigned int n_rules;
1203 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1204 simap_increase(usage, "ops",
1205 ofproto->n_pending + hmap_count(&ofproto->deletions));
1208 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1209 n_rules += classifier_count(&table->cls);
1211 simap_increase(usage, "rules", n_rules);
1213 if (ofproto->ofproto_class->get_memory_usage) {
1214 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1217 connmgr_get_memory_usage(ofproto->connmgr, usage);
1221 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1224 connmgr_get_controller_info(ofproto->connmgr, info);
1228 ofproto_free_ofproto_controller_info(struct shash *info)
1230 connmgr_free_controller_info(info);
1233 /* Makes a deep copy of 'old' into 'port'. */
1235 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1237 port->name = xstrdup(old->name);
1238 port->type = xstrdup(old->type);
1239 port->ofp_port = old->ofp_port;
1242 /* Frees memory allocated to members of 'ofproto_port'.
1244 * Do not call this function on an ofproto_port obtained from
1245 * ofproto_port_dump_next(): that function retains ownership of the data in the
1248 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1250 free(ofproto_port->name);
1251 free(ofproto_port->type);
1254 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1256 * This function provides no status indication. An error status for the entire
1257 * dump operation is provided when it is completed by calling
1258 * ofproto_port_dump_done().
1261 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1262 const struct ofproto *ofproto)
1264 dump->ofproto = ofproto;
1265 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1269 /* Attempts to retrieve another port from 'dump', which must have been created
1270 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1271 * 'port' and returns true. On failure, returns false.
1273 * Failure might indicate an actual error or merely that the last port has been
1274 * dumped. An error status for the entire dump operation is provided when it
1275 * is completed by calling ofproto_port_dump_done().
1277 * The ofproto owns the data stored in 'port'. It will remain valid until at
1278 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1279 * ofproto_port_dump_done(). */
1281 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1282 struct ofproto_port *port)
1284 const struct ofproto *ofproto = dump->ofproto;
1290 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1293 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1299 /* Completes port table dump operation 'dump', which must have been created
1300 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1301 * error-free, otherwise a positive errno value describing the problem. */
1303 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1305 const struct ofproto *ofproto = dump->ofproto;
1307 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1310 return dump->error == EOF ? 0 : dump->error;
1313 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1314 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1315 * is non-null). On failure, returns a positive errno value and sets
1316 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1318 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1319 uint16_t *ofp_portp)
1324 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1326 update_port(ofproto, netdev_get_name(netdev));
1329 *ofp_portp = error ? OFPP_NONE : ofp_port;
1334 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1335 * initializes '*port' appropriately; on failure, returns a positive errno
1338 * The caller owns the data in 'ofproto_port' and must free it with
1339 * ofproto_port_destroy() when it is no longer needed. */
1341 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1342 struct ofproto_port *port)
1346 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1348 memset(port, 0, sizeof *port);
1353 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1354 * Returns 0 if successful, otherwise a positive errno. */
1356 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1358 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1359 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1362 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1363 if (!error && ofport) {
1364 /* 'name' is the netdev's name and update_port() is going to close the
1365 * netdev. Just in case update_port() refers to 'name' after it
1366 * destroys 'ofport', make a copy of it around the update_port()
1368 char *devname = xstrdup(name);
1369 update_port(ofproto, devname);
1375 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1376 * performs the 'n_actions' actions in 'actions'. The new flow will not
1379 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1380 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1381 * controllers; otherwise, it will be hidden.
1383 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1385 * This is a helper function for in-band control and fail-open. */
1387 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1388 const struct ofpact *ofpacts, size_t ofpacts_len)
1390 const struct rule *rule;
1392 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1393 &ofproto->tables[0].cls, cls_rule));
1394 if (!rule || !ofpacts_equal(rule->ofpacts, rule->ofpacts_len,
1395 ofpacts, ofpacts_len)) {
1396 struct ofputil_flow_mod fm;
1398 memset(&fm, 0, sizeof fm);
1400 fm.buffer_id = UINT32_MAX;
1401 fm.ofpacts = xmemdup(ofpacts, ofpacts_len);
1402 fm.ofpacts_len = ofpacts_len;
1403 add_flow(ofproto, NULL, &fm, NULL);
1408 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1409 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1410 * operation cannot be initiated now but may be retried later.
1412 * This is a helper function for in-band control and fail-open. */
1414 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1416 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1419 /* Searches for a rule with matching criteria exactly equal to 'target' in
1420 * ofproto's table 0 and, if it finds one, deletes it.
1422 * This is a helper function for in-band control and fail-open. */
1424 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1428 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1429 &ofproto->tables[0].cls, target));
1431 /* No such rule -> success. */
1433 } else if (rule->pending) {
1434 /* An operation on the rule is already pending -> failure.
1435 * Caller must retry later if it's important. */
1438 /* Initiate deletion -> success. */
1439 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1440 ofoperation_create(group, rule, OFOPERATION_DELETE);
1441 oftable_remove_rule(rule);
1442 ofproto->ofproto_class->rule_destruct(rule);
1443 ofopgroup_submit(group);
1449 /* Starts the process of deleting all of the flows from all of ofproto's flow
1450 * tables and then reintroducing the flows required by in-band control and
1451 * fail-open. The process will complete in a later call to ofproto_run(). */
1453 ofproto_flush_flows(struct ofproto *ofproto)
1455 COVERAGE_INC(ofproto_flush);
1456 ofproto->state = S_FLUSH;
1460 reinit_ports(struct ofproto *p)
1462 struct ofproto_port_dump dump;
1463 struct sset devnames;
1464 struct ofport *ofport;
1465 struct ofproto_port ofproto_port;
1466 const char *devname;
1468 COVERAGE_INC(ofproto_reinit_ports);
1470 sset_init(&devnames);
1471 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1472 sset_add(&devnames, netdev_get_name(ofport->netdev));
1474 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1475 sset_add(&devnames, ofproto_port.name);
1478 SSET_FOR_EACH (devname, &devnames) {
1479 update_port(p, devname);
1481 sset_destroy(&devnames);
1484 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
1485 * pointer if the netdev cannot be opened. On success, also fills in
1487 static struct netdev *
1488 ofport_open(const struct ofproto *ofproto,
1489 const struct ofproto_port *ofproto_port,
1490 struct ofputil_phy_port *pp)
1492 enum netdev_flags flags;
1493 struct netdev *netdev;
1496 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1498 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
1499 "cannot be opened (%s)",
1501 ofproto_port->name, ofproto_port->ofp_port,
1502 ofproto_port->name, strerror(error));
1506 pp->port_no = ofproto_port->ofp_port;
1507 netdev_get_etheraddr(netdev, pp->hw_addr);
1508 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
1509 netdev_get_flags(netdev, &flags);
1510 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
1511 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
1512 netdev_get_features(netdev, &pp->curr, &pp->advertised,
1513 &pp->supported, &pp->peer);
1514 pp->curr_speed = netdev_features_to_bps(pp->curr);
1515 pp->max_speed = netdev_features_to_bps(pp->supported);
1520 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1521 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
1524 ofport_equal(const struct ofputil_phy_port *a,
1525 const struct ofputil_phy_port *b)
1527 return (eth_addr_equals(a->hw_addr, b->hw_addr)
1528 && a->state == b->state
1529 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
1530 && a->curr == b->curr
1531 && a->advertised == b->advertised
1532 && a->supported == b->supported
1533 && a->peer == b->peer
1534 && a->curr_speed == b->curr_speed
1535 && a->max_speed == b->max_speed);
1538 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1539 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1540 * one with the same name or port number). */
1542 ofport_install(struct ofproto *p,
1543 struct netdev *netdev, const struct ofputil_phy_port *pp)
1545 const char *netdev_name = netdev_get_name(netdev);
1546 struct ofport *ofport;
1549 /* Create ofport. */
1550 ofport = p->ofproto_class->port_alloc();
1555 ofport->ofproto = p;
1556 ofport->netdev = netdev;
1557 ofport->change_seq = netdev_change_seq(netdev);
1559 ofport->ofp_port = pp->port_no;
1561 /* Add port to 'p'. */
1562 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1563 shash_add(&p->port_by_name, netdev_name, ofport);
1565 update_mtu(p, ofport);
1567 /* Let the ofproto_class initialize its private data. */
1568 error = p->ofproto_class->port_construct(ofport);
1572 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
1576 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1577 p->name, netdev_name, strerror(error));
1579 ofport_destroy__(ofport);
1581 netdev_close(netdev);
1585 /* Removes 'ofport' from 'p' and destroys it. */
1587 ofport_remove(struct ofport *ofport)
1589 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
1591 ofport_destroy(ofport);
1594 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1597 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1599 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1601 ofport_remove(port);
1605 /* Updates 'port' with new 'pp' description.
1607 * Does not handle a name or port number change. The caller must implement
1608 * such a change as a delete followed by an add. */
1610 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
1612 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
1613 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
1614 | (pp->config & OFPUTIL_PC_PORT_DOWN));
1615 port->pp.state = pp->state;
1616 port->pp.curr = pp->curr;
1617 port->pp.advertised = pp->advertised;
1618 port->pp.supported = pp->supported;
1619 port->pp.peer = pp->peer;
1620 port->pp.curr_speed = pp->curr_speed;
1621 port->pp.max_speed = pp->max_speed;
1623 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
1626 /* Update OpenFlow 'state' in 'port' and notify controller. */
1628 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
1630 if (port->pp.state != state) {
1631 port->pp.state = state;
1632 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
1638 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1640 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1642 if (port->ofproto->ofproto_class->set_realdev) {
1643 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1645 if (port->ofproto->ofproto_class->set_stp_port) {
1646 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1648 if (port->ofproto->ofproto_class->set_cfm) {
1649 port->ofproto->ofproto_class->set_cfm(port, NULL);
1651 if (port->ofproto->ofproto_class->bundle_remove) {
1652 port->ofproto->ofproto_class->bundle_remove(port);
1658 ofport_destroy__(struct ofport *port)
1660 struct ofproto *ofproto = port->ofproto;
1661 const char *name = netdev_get_name(port->netdev);
1663 hmap_remove(&ofproto->ports, &port->hmap_node);
1664 shash_delete(&ofproto->port_by_name,
1665 shash_find(&ofproto->port_by_name, name));
1667 netdev_close(port->netdev);
1668 ofproto->ofproto_class->port_dealloc(port);
1672 ofport_destroy(struct ofport *port)
1675 port->ofproto->ofproto_class->port_destruct(port);
1676 ofport_destroy__(port);
1681 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1683 struct ofport *port;
1685 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1686 hash_int(ofp_port, 0), &ofproto->ports) {
1687 if (port->ofp_port == ofp_port) {
1695 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
1697 struct ofproto *ofproto = port->ofproto;
1700 if (ofproto->ofproto_class->port_get_stats) {
1701 error = ofproto->ofproto_class->port_get_stats(port, stats);
1710 update_port(struct ofproto *ofproto, const char *name)
1712 struct ofproto_port ofproto_port;
1713 struct ofputil_phy_port pp;
1714 struct netdev *netdev;
1715 struct ofport *port;
1717 COVERAGE_INC(ofproto_update_port);
1719 /* Fetch 'name''s location and properties from the datapath. */
1720 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1721 ? ofport_open(ofproto, &ofproto_port, &pp)
1724 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1725 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1726 struct netdev *old_netdev = port->netdev;
1728 /* 'name' hasn't changed location. Any properties changed? */
1729 if (!ofport_equal(&port->pp, &pp)) {
1730 ofport_modified(port, &pp);
1733 update_mtu(ofproto, port);
1735 /* Install the newly opened netdev in case it has changed.
1736 * Don't close the old netdev yet in case port_modified has to
1737 * remove a retained reference to it.*/
1738 port->netdev = netdev;
1739 port->change_seq = netdev_change_seq(netdev);
1741 if (port->ofproto->ofproto_class->port_modified) {
1742 port->ofproto->ofproto_class->port_modified(port);
1745 netdev_close(old_netdev);
1747 /* If 'port' is nonnull then its name differs from 'name' and thus
1748 * we should delete it. If we think there's a port named 'name'
1749 * then its port number must be wrong now so delete it too. */
1751 ofport_remove(port);
1753 ofport_remove_with_name(ofproto, name);
1754 ofport_install(ofproto, netdev, &pp);
1757 /* Any port named 'name' is gone now. */
1758 ofport_remove_with_name(ofproto, name);
1760 ofproto_port_destroy(&ofproto_port);
1764 init_ports(struct ofproto *p)
1766 struct ofproto_port_dump dump;
1767 struct ofproto_port ofproto_port;
1769 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1770 uint16_t ofp_port = ofproto_port.ofp_port;
1771 if (ofproto_get_port(p, ofp_port)) {
1772 VLOG_WARN_RL(&rl, "%s: ignoring duplicate port %"PRIu16" "
1773 "in datapath", p->name, ofp_port);
1774 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1775 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
1776 p->name, ofproto_port.name);
1778 struct ofputil_phy_port pp;
1779 struct netdev *netdev;
1781 netdev = ofport_open(p, &ofproto_port, &pp);
1783 ofport_install(p, netdev, &pp);
1791 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1792 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1794 find_min_mtu(struct ofproto *p)
1796 struct ofport *ofport;
1799 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1800 struct netdev *netdev = ofport->netdev;
1803 /* Skip any internal ports, since that's what we're trying to
1805 if (!strcmp(netdev_get_type(netdev), "internal")) {
1809 if (netdev_get_mtu(netdev, &dev_mtu)) {
1812 if (!mtu || dev_mtu < mtu) {
1817 return mtu ? mtu: ETH_PAYLOAD_MAX;
1820 /* Update MTU of all datapath devices on 'p' to the minimum of the
1821 * non-datapath ports in event of 'port' added or changed. */
1823 update_mtu(struct ofproto *p, struct ofport *port)
1825 struct ofport *ofport;
1826 struct netdev *netdev = port->netdev;
1827 int dev_mtu, old_min;
1829 if (netdev_get_mtu(netdev, &dev_mtu)) {
1833 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
1834 if (dev_mtu > p->min_mtu) {
1835 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
1836 dev_mtu = p->min_mtu;
1839 port->mtu = dev_mtu;
1843 /* For non-internal port find new min mtu. */
1844 old_min = p->min_mtu;
1845 port->mtu = dev_mtu;
1846 p->min_mtu = find_min_mtu(p);
1847 if (p->min_mtu == old_min) {
1851 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1852 struct netdev *netdev = ofport->netdev;
1854 if (!strcmp(netdev_get_type(netdev), "internal")) {
1855 if (!netdev_set_mtu(netdev, p->min_mtu)) {
1856 ofport->mtu = p->min_mtu;
1863 ofproto_rule_destroy__(struct rule *rule)
1866 free(rule->ofpacts);
1867 rule->ofproto->ofproto_class->rule_dealloc(rule);
1871 /* This function allows an ofproto implementation to destroy any rules that
1872 * remain when its ->destruct() function is called. The caller must have
1873 * already uninitialized any derived members of 'rule' (step 5 described in the
1874 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1875 * This function implements steps 6 and 7.
1877 * This function should only be called from an ofproto implementation's
1878 * ->destruct() function. It is not suitable elsewhere. */
1880 ofproto_rule_destroy(struct rule *rule)
1882 assert(!rule->pending);
1883 oftable_remove_rule(rule);
1884 ofproto_rule_destroy__(rule);
1887 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1888 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
1890 rule_has_out_port(const struct rule *rule, uint16_t port)
1892 return (port == OFPP_NONE
1893 || ofpacts_output_to_port(rule->ofpacts, rule->ofpacts_len, port));
1896 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1897 * statistics appropriately. 'packet' must have at least sizeof(struct
1898 * ofp_packet_in) bytes of headroom.
1900 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1901 * with statistics for 'packet' either way.
1903 * Takes ownership of 'packet'. */
1905 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1909 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1911 flow_extract(packet, 0, 0, in_port, &flow);
1912 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1915 /* Returns true if 'rule' should be hidden from the controller.
1917 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1918 * (e.g. by in-band control) and are intentionally hidden from the
1921 rule_is_hidden(const struct rule *rule)
1923 return rule->cr.priority > UINT16_MAX;
1926 static enum oftable_flags
1927 rule_get_flags(const struct rule *rule)
1929 return rule->ofproto->tables[rule->table_id].flags;
1933 rule_is_modifiable(const struct rule *rule)
1935 return !(rule_get_flags(rule) & OFTABLE_READONLY);
1939 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1941 ofconn_send_reply(ofconn, make_echo_reply(oh));
1946 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1948 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1949 struct ofputil_switch_features features;
1950 struct ofport *port;
1954 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
1956 assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
1958 features.datapath_id = ofproto->datapath_id;
1959 features.n_buffers = pktbuf_capacity();
1960 features.n_tables = ofproto->n_tables;
1961 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
1962 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
1964 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
1967 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
1969 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1970 ofputil_put_switch_features_port(&port->pp, b);
1973 ofconn_send_reply(ofconn, b);
1978 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1980 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1981 struct ofp_switch_config *osc;
1982 enum ofp_config_flags flags;
1986 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1987 flags = ofproto->frag_handling;
1988 if (ofconn_get_invalid_ttl_to_controller(ofconn)) {
1989 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
1991 osc->flags = htons(flags);
1992 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1993 ofconn_send_reply(ofconn, buf);
1999 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
2001 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2002 uint16_t flags = ntohs(osc->flags);
2004 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2005 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
2006 enum ofp_config_flags cur = ofproto->frag_handling;
2007 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2009 assert((cur & OFPC_FRAG_MASK) == cur);
2011 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2012 ofproto->frag_handling = next;
2014 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2016 ofputil_frag_handling_to_string(next));
2020 ofconn_set_invalid_ttl_to_controller(ofconn,
2021 (flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2023 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2028 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2029 * error message code for the caller to propagate upward. Otherwise, returns
2032 * The log message mentions 'msg_type'. */
2034 reject_slave_controller(struct ofconn *ofconn)
2036 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2037 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
2038 return OFPERR_OFPBRC_EPERM;
2045 handle_packet_out(struct ofconn *ofconn, const struct ofp_packet_out *opo)
2047 struct ofproto *p = ofconn_get_ofproto(ofconn);
2048 struct ofputil_packet_out po;
2049 struct ofpbuf *payload;
2050 uint64_t ofpacts_stub[1024 / 8];
2051 struct ofpbuf ofpacts;
2055 COVERAGE_INC(ofproto_packet_out);
2057 error = reject_slave_controller(ofconn);
2062 /* Decode message. */
2063 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2064 error = ofputil_decode_packet_out(&po, opo, &ofpacts);
2066 goto exit_free_ofpacts;
2070 if (po.buffer_id != UINT32_MAX) {
2071 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2072 if (error || !payload) {
2073 goto exit_free_ofpacts;
2076 payload = xmalloc(sizeof *payload);
2077 ofpbuf_use_const(payload, po.packet, po.packet_len);
2080 /* Send out packet. */
2081 flow_extract(payload, 0, 0, po.in_port, &flow);
2082 error = p->ofproto_class->packet_out(p, payload, &flow,
2083 po.ofpacts, po.ofpacts_len);
2084 ofpbuf_delete(payload);
2087 ofpbuf_uninit(&ofpacts);
2093 update_port_config(struct ofport *port,
2094 enum ofputil_port_config config,
2095 enum ofputil_port_config mask)
2097 enum ofputil_port_config old_config = port->pp.config;
2098 enum ofputil_port_config toggle;
2100 toggle = (config ^ port->pp.config) & mask;
2101 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2102 if (config & OFPUTIL_PC_PORT_DOWN) {
2103 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
2105 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
2107 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2110 port->pp.config ^= toggle;
2111 if (port->pp.config != old_config) {
2112 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2117 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2119 struct ofproto *p = ofconn_get_ofproto(ofconn);
2120 struct ofputil_port_mod pm;
2121 struct ofport *port;
2124 error = reject_slave_controller(ofconn);
2129 error = ofputil_decode_port_mod(oh, &pm);
2134 port = ofproto_get_port(p, pm.port_no);
2136 return OFPERR_OFPPMFC_BAD_PORT;
2137 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2138 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2140 update_port_config(port, pm.config, pm.mask);
2142 netdev_set_advertisements(port->netdev, pm.advertise);
2149 handle_desc_stats_request(struct ofconn *ofconn,
2150 const struct ofp_stats_msg *request)
2152 struct ofproto *p = ofconn_get_ofproto(ofconn);
2153 struct ofp_desc_stats *ods;
2156 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
2157 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
2158 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
2159 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
2160 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
2161 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
2162 ofconn_send_reply(ofconn, msg);
2168 handle_table_stats_request(struct ofconn *ofconn,
2169 const struct ofp_stats_msg *request)
2171 struct ofproto *p = ofconn_get_ofproto(ofconn);
2172 struct ofp_table_stats *ots;
2176 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
2178 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
2179 for (i = 0; i < p->n_tables; i++) {
2180 ots[i].table_id = i;
2181 sprintf(ots[i].name, "table%zu", i);
2182 ots[i].wildcards = htonl(OFPFW10_ALL);
2183 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2184 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
2187 p->ofproto_class->get_tables(p, ots);
2189 for (i = 0; i < p->n_tables; i++) {
2190 const struct oftable *table = &p->tables[i];
2193 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
2196 if (table->max_flows < ntohl(ots[i].max_entries)) {
2197 ots[i].max_entries = htonl(table->max_flows);
2201 ofconn_send_reply(ofconn, msg);
2206 append_port_stat(struct ofport *port, struct list *replies)
2208 struct netdev_stats stats;
2209 struct ofp_port_stats *ops;
2211 /* Intentionally ignore return value, since errors will set
2212 * 'stats' to all-1s, which is correct for OpenFlow, and
2213 * netdev_get_stats() will log errors. */
2214 ofproto_port_get_stats(port, &stats);
2216 ops = ofputil_append_stats_reply(sizeof *ops, replies);
2217 ops->port_no = htons(port->pp.port_no);
2218 memset(ops->pad, 0, sizeof ops->pad);
2219 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
2220 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
2221 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
2222 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
2223 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
2224 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
2225 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
2226 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
2227 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
2228 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
2229 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
2230 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
2234 handle_port_stats_request(struct ofconn *ofconn,
2235 const struct ofp_port_stats_request *psr)
2237 struct ofproto *p = ofconn_get_ofproto(ofconn);
2238 struct ofport *port;
2239 struct list replies;
2241 ofputil_start_stats_reply(&psr->osm, &replies);
2242 if (psr->port_no != htons(OFPP_NONE)) {
2243 port = ofproto_get_port(p, ntohs(psr->port_no));
2245 append_port_stat(port, &replies);
2248 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2249 append_port_stat(port, &replies);
2253 ofconn_send_replies(ofconn, &replies);
2258 handle_port_desc_stats_request(struct ofconn *ofconn,
2259 const struct ofp_stats_msg *osm)
2261 struct ofproto *p = ofconn_get_ofproto(ofconn);
2262 struct ofport *port;
2263 struct list replies;
2265 ofputil_start_stats_reply(osm, &replies);
2267 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2268 ofputil_append_port_desc_stats_reply(ofconn_get_protocol(ofconn),
2269 &port->pp, &replies);
2272 ofconn_send_replies(ofconn, &replies);
2277 calc_flow_duration__(long long int start, long long int now,
2278 uint32_t *sec, uint32_t *nsec)
2280 long long int msecs = now - start;
2281 *sec = msecs / 1000;
2282 *nsec = (msecs % 1000) * (1000 * 1000);
2285 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2286 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2288 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2290 return (table_id == 0xff || table_id < ofproto->n_tables
2292 : OFPERR_NXBRC_BAD_TABLE_ID);
2296 static struct oftable *
2297 next_visible_table(struct ofproto *ofproto, uint8_t table_id)
2299 struct oftable *table;
2301 for (table = &ofproto->tables[table_id];
2302 table < &ofproto->tables[ofproto->n_tables];
2304 if (!(table->flags & OFTABLE_HIDDEN)) {
2312 static struct oftable *
2313 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2315 if (table_id == 0xff) {
2316 return next_visible_table(ofproto, 0);
2317 } else if (table_id < ofproto->n_tables) {
2318 return &ofproto->tables[table_id];
2324 static struct oftable *
2325 next_matching_table(struct ofproto *ofproto,
2326 struct oftable *table, uint8_t table_id)
2328 return (table_id == 0xff
2329 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
2333 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
2335 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2336 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
2338 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2339 * only once, for that table. (This can be used to access tables marked
2342 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2343 * entered at all. (Perhaps you should have validated TABLE_ID with
2344 * check_table_id().)
2346 * All parameters are evaluated multiple times.
2348 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
2349 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
2351 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
2353 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2354 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2355 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2358 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2359 * to 'out_port' are included.
2361 * Hidden rules are always omitted.
2363 * Returns 0 on success, otherwise an OpenFlow error code. */
2365 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2366 const struct cls_rule *match,
2367 ovs_be64 cookie, ovs_be64 cookie_mask,
2368 uint16_t out_port, struct list *rules)
2370 struct oftable *table;
2373 error = check_table_id(ofproto, table_id);
2379 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2380 struct cls_cursor cursor;
2383 cls_cursor_init(&cursor, &table->cls, match);
2384 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2385 if (rule->pending) {
2386 return OFPROTO_POSTPONE;
2388 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2389 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2390 list_push_back(rules, &rule->ofproto_node);
2397 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2398 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2399 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2402 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2403 * to 'out_port' are included.
2405 * Hidden rules are always omitted.
2407 * Returns 0 on success, otherwise an OpenFlow error code. */
2409 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2410 const struct cls_rule *match,
2411 ovs_be64 cookie, ovs_be64 cookie_mask,
2412 uint16_t out_port, struct list *rules)
2414 struct oftable *table;
2417 error = check_table_id(ofproto, table_id);
2423 FOR_EACH_MATCHING_TABLE (table, table_id, ofproto) {
2426 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls,
2429 if (rule->pending) {
2430 return OFPROTO_POSTPONE;
2432 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)
2433 && !((rule->flow_cookie ^ cookie) & cookie_mask)) {
2434 list_push_back(rules, &rule->ofproto_node);
2441 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
2442 * forced into the range of a uint16_t. */
2444 age_secs(long long int age_ms)
2446 return (age_ms < 0 ? 0
2447 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
2448 : (unsigned int) age_ms / 1000);
2452 handle_flow_stats_request(struct ofconn *ofconn,
2453 const struct ofp_stats_msg *osm)
2455 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2456 struct ofputil_flow_stats_request fsr;
2457 struct list replies;
2462 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2467 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2468 fsr.cookie, fsr.cookie_mask,
2469 fsr.out_port, &rules);
2474 ofputil_start_stats_reply(osm, &replies);
2475 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2476 long long int now = time_msec();
2477 struct ofputil_flow_stats fs;
2480 fs.cookie = rule->flow_cookie;
2481 fs.table_id = rule->table_id;
2482 calc_flow_duration__(rule->created, now, &fs.duration_sec,
2484 fs.idle_timeout = rule->idle_timeout;
2485 fs.hard_timeout = rule->hard_timeout;
2486 fs.idle_age = age_secs(now - rule->used);
2487 fs.hard_age = age_secs(now - rule->modified);
2488 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2490 fs.ofpacts = rule->ofpacts;
2491 fs.ofpacts_len = rule->ofpacts_len;
2492 ofputil_append_flow_stats_reply(&fs, &replies);
2494 ofconn_send_replies(ofconn, &replies);
2500 flow_stats_ds(struct rule *rule, struct ds *results)
2502 uint64_t packet_count, byte_count;
2504 rule->ofproto->ofproto_class->rule_get_stats(rule,
2505 &packet_count, &byte_count);
2507 if (rule->table_id != 0) {
2508 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2510 ds_put_format(results, "duration=%llds, ",
2511 (time_msec() - rule->created) / 1000);
2512 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2513 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2514 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2515 cls_rule_format(&rule->cr, results);
2516 ds_put_char(results, ',');
2517 if (rule->ofpacts_len > 0) {
2518 ofpacts_format(rule->ofpacts, rule->ofpacts_len, results);
2520 ds_put_cstr(results, "drop");
2522 ds_put_cstr(results, "\n");
2525 /* Adds a pretty-printed description of all flows to 'results', including
2526 * hidden flows (e.g., set up by in-band control). */
2528 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2530 struct oftable *table;
2532 OFPROTO_FOR_EACH_TABLE (table, p) {
2533 struct cls_cursor cursor;
2536 cls_cursor_init(&cursor, &table->cls, NULL);
2537 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2538 flow_stats_ds(rule, results);
2543 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2544 * '*engine_type' and '*engine_id', respectively. */
2546 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2547 uint8_t *engine_type, uint8_t *engine_id)
2549 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2552 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns a
2553 * bitmask of 'cfm_fault_reason's to indicate a CFM fault (generally
2554 * indicating a connectivity problem). Returns zero if CFM is not faulted,
2555 * and -1 if CFM is not enabled on 'port'. */
2557 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2559 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2560 return (ofport && ofproto->ofproto_class->get_cfm_fault
2561 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2565 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2566 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2567 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2568 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2570 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2571 uint16_t ofp_port, const uint64_t **rmps,
2574 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2578 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2579 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2584 /* Checks the health of the CFM for 'ofp_port' within 'ofproto'. Returns an
2585 * integer value between 0 and 100 to indicate the health of the port as a
2586 * percentage which is the average of cfm health of all the remote_mpids or
2587 * returns -1 if CFM is not enabled on 'ofport'. */
2589 ofproto_port_get_cfm_health(const struct ofproto *ofproto, uint16_t ofp_port)
2591 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2592 return (ofport && ofproto->ofproto_class->get_cfm_health
2593 ? ofproto->ofproto_class->get_cfm_health(ofport)
2598 handle_aggregate_stats_request(struct ofconn *ofconn,
2599 const struct ofp_stats_msg *osm)
2601 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2602 struct ofputil_flow_stats_request request;
2603 struct ofputil_aggregate_stats stats;
2604 bool unknown_packets, unknown_bytes;
2605 struct ofpbuf *reply;
2610 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2615 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2616 request.cookie, request.cookie_mask,
2617 request.out_port, &rules);
2622 memset(&stats, 0, sizeof stats);
2623 unknown_packets = unknown_bytes = false;
2624 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2625 uint64_t packet_count;
2626 uint64_t byte_count;
2628 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2631 if (packet_count == UINT64_MAX) {
2632 unknown_packets = true;
2634 stats.packet_count += packet_count;
2637 if (byte_count == UINT64_MAX) {
2638 unknown_bytes = true;
2640 stats.byte_count += byte_count;
2645 if (unknown_packets) {
2646 stats.packet_count = UINT64_MAX;
2648 if (unknown_bytes) {
2649 stats.byte_count = UINT64_MAX;
2652 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2653 ofconn_send_reply(ofconn, reply);
2658 struct queue_stats_cbdata {
2659 struct ofport *ofport;
2660 struct list replies;
2664 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2665 const struct netdev_queue_stats *stats)
2667 struct ofp_queue_stats *reply;
2669 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2670 reply->port_no = htons(cbdata->ofport->pp.port_no);
2671 memset(reply->pad, 0, sizeof reply->pad);
2672 reply->queue_id = htonl(queue_id);
2673 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2674 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2675 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2679 handle_queue_stats_dump_cb(uint32_t queue_id,
2680 struct netdev_queue_stats *stats,
2683 struct queue_stats_cbdata *cbdata = cbdata_;
2685 put_queue_stats(cbdata, queue_id, stats);
2689 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2690 struct queue_stats_cbdata *cbdata)
2692 cbdata->ofport = port;
2693 if (queue_id == OFPQ_ALL) {
2694 netdev_dump_queue_stats(port->netdev,
2695 handle_queue_stats_dump_cb, cbdata);
2697 struct netdev_queue_stats stats;
2699 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2700 put_queue_stats(cbdata, queue_id, &stats);
2702 return OFPERR_OFPQOFC_BAD_QUEUE;
2709 handle_queue_stats_request(struct ofconn *ofconn,
2710 const struct ofp_queue_stats_request *qsr)
2712 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2713 struct queue_stats_cbdata cbdata;
2714 unsigned int port_no;
2715 struct ofport *port;
2719 COVERAGE_INC(ofproto_queue_req);
2721 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2723 port_no = ntohs(qsr->port_no);
2724 queue_id = ntohl(qsr->queue_id);
2725 if (port_no == OFPP_ALL) {
2726 error = OFPERR_OFPQOFC_BAD_QUEUE;
2727 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2728 if (!handle_queue_stats_for_port(port, queue_id, &cbdata)) {
2733 port = ofproto_get_port(ofproto, port_no);
2735 ? handle_queue_stats_for_port(port, queue_id, &cbdata)
2736 : OFPERR_OFPQOFC_BAD_PORT);
2739 ofconn_send_replies(ofconn, &cbdata.replies);
2741 ofpbuf_list_delete(&cbdata.replies);
2748 is_flow_deletion_pending(const struct ofproto *ofproto,
2749 const struct cls_rule *cls_rule,
2752 if (!hmap_is_empty(&ofproto->deletions)) {
2753 struct ofoperation *op;
2755 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2756 cls_rule_hash(cls_rule, table_id),
2757 &ofproto->deletions) {
2758 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2767 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2768 * in which no matching flow already exists in the flow table.
2770 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2771 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2772 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
2773 * initiated now but may be retried later.
2775 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
2776 * ownership remains with the caller.
2778 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2781 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2782 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2784 struct oftable *table;
2785 struct ofopgroup *group;
2786 struct rule *victim;
2790 error = check_table_id(ofproto, fm->table_id);
2796 if (fm->table_id == 0xff) {
2798 if (ofproto->ofproto_class->rule_choose_table) {
2799 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2804 assert(table_id < ofproto->n_tables);
2805 table = &ofproto->tables[table_id];
2807 table = &ofproto->tables[0];
2809 } else if (fm->table_id < ofproto->n_tables) {
2810 table = &ofproto->tables[fm->table_id];
2812 return OFPERR_NXFMFC_BAD_TABLE_ID;
2815 if (table->flags & OFTABLE_READONLY) {
2816 return OFPERR_OFPBRC_EPERM;
2819 /* Check for overlap, if requested. */
2820 if (fm->flags & OFPFF_CHECK_OVERLAP
2821 && classifier_rule_overlaps(&table->cls, &fm->cr)) {
2822 return OFPERR_OFPFMFC_OVERLAP;
2825 /* Serialize against pending deletion. */
2826 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2827 return OFPROTO_POSTPONE;
2830 /* Allocate new rule. */
2831 rule = ofproto->ofproto_class->rule_alloc();
2833 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2834 ofproto->name, strerror(error));
2837 rule->ofproto = ofproto;
2839 rule->pending = NULL;
2840 rule->flow_cookie = fm->new_cookie;
2841 rule->created = rule->modified = rule->used = time_msec();
2842 rule->idle_timeout = fm->idle_timeout;
2843 rule->hard_timeout = fm->hard_timeout;
2844 rule->table_id = table - ofproto->tables;
2845 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2846 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2847 rule->ofpacts_len = fm->ofpacts_len;
2848 rule->evictable = true;
2849 rule->eviction_group = NULL;
2851 /* Insert new rule. */
2852 victim = oftable_replace_rule(rule);
2853 if (victim && !rule_is_modifiable(victim)) {
2854 error = OFPERR_OFPBRC_EPERM;
2855 } else if (victim && victim->pending) {
2856 error = OFPROTO_POSTPONE;
2860 if (classifier_count(&table->cls) > table->max_flows) {
2863 was_evictable = rule->evictable;
2864 rule->evictable = false;
2865 evict = choose_rule_to_evict(table);
2866 rule->evictable = was_evictable;
2869 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
2871 } else if (evict->pending) {
2872 error = OFPROTO_POSTPONE;
2879 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2880 ofoperation_create(group, rule, OFOPERATION_ADD);
2881 rule->pending->victim = victim;
2883 error = ofproto->ofproto_class->rule_construct(rule);
2885 ofoperation_destroy(rule->pending);
2887 delete_flow__(evict, group);
2889 ofopgroup_submit(group);
2893 /* Back out if an error occurred. */
2895 oftable_substitute_rule(rule, victim);
2896 ofproto_rule_destroy__(rule);
2901 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2903 /* Modifies the rules listed in 'rules', changing their actions to match those
2906 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2909 * Returns 0 on success, otherwise an OpenFlow error code. */
2911 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2912 const struct ofputil_flow_mod *fm,
2913 const struct ofp_header *request, struct list *rules)
2915 struct ofopgroup *group;
2919 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2920 error = OFPERR_OFPBRC_EPERM;
2921 LIST_FOR_EACH (rule, ofproto_node, rules) {
2922 if (rule_is_modifiable(rule)) {
2923 /* At least one rule is modifiable, don't report EPERM error. */
2929 if (!ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2930 rule->ofpacts, rule->ofpacts_len)) {
2931 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2932 rule->pending->ofpacts = rule->ofpacts;
2933 rule->pending->ofpacts_len = rule->ofpacts_len;
2934 rule->ofpacts = xmemdup(fm->ofpacts, fm->ofpacts_len);
2935 rule->ofpacts_len = fm->ofpacts_len;
2936 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2938 rule->modified = time_msec();
2940 if (fm->new_cookie != htonll(UINT64_MAX)) {
2941 rule->flow_cookie = fm->new_cookie;
2944 ofopgroup_submit(group);
2949 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
2952 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2955 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2956 const struct ofputil_flow_mod *fm,
2957 const struct ofp_header *request)
2962 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
2963 fm->cookie, fm->cookie_mask,
2967 } else if (list_is_empty(&rules)) {
2968 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2970 return modify_flows__(ofproto, ofconn, fm, request, &rules);
2974 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2977 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2980 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2981 const struct ofputil_flow_mod *fm,
2982 const struct ofp_header *request)
2987 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
2988 fm->cookie, fm->cookie_mask,
2993 } else if (list_is_empty(&rules)) {
2994 return fm->cookie_mask ? 0 : add_flow(ofproto, ofconn, fm, request);
2996 return list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2997 fm, request, &rules)
3002 /* OFPFC_DELETE implementation. */
3005 delete_flow__(struct rule *rule, struct ofopgroup *group)
3007 struct ofproto *ofproto = rule->ofproto;
3009 ofproto_rule_send_removed(rule, OFPRR_DELETE);
3011 ofoperation_create(group, rule, OFOPERATION_DELETE);
3012 oftable_remove_rule(rule);
3013 ofproto->ofproto_class->rule_destruct(rule);
3016 /* Deletes the rules listed in 'rules'.
3018 * Returns 0 on success, otherwise an OpenFlow error code. */
3020 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
3021 const struct ofp_header *request, struct list *rules)
3023 struct rule *rule, *next;
3024 struct ofopgroup *group;
3026 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
3027 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
3028 delete_flow__(rule, group);
3030 ofopgroup_submit(group);
3035 /* Implements OFPFC_DELETE. */
3037 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
3038 const struct ofputil_flow_mod *fm,
3039 const struct ofp_header *request)
3044 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr,
3045 fm->cookie, fm->cookie_mask,
3046 fm->out_port, &rules);
3047 return (error ? error
3048 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
3053 /* Implements OFPFC_DELETE_STRICT. */
3055 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
3056 const struct ofputil_flow_mod *fm,
3057 const struct ofp_header *request)
3062 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr,
3063 fm->cookie, fm->cookie_mask,
3064 fm->out_port, &rules);
3065 return (error ? error
3066 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
3072 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
3074 struct ofputil_flow_removed fr;
3076 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
3081 fr.cookie = rule->flow_cookie;
3083 calc_flow_duration__(rule->created, time_msec(),
3084 &fr.duration_sec, &fr.duration_nsec);
3085 fr.idle_timeout = rule->idle_timeout;
3086 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
3089 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
3093 ofproto_rule_update_used(struct rule *rule, long long int used)
3095 if (used > rule->used) {
3096 struct eviction_group *evg = rule->eviction_group;
3100 heap_change(&evg->rules, &rule->evg_node,
3101 rule_eviction_priority(rule));
3106 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
3107 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
3110 * ofproto implementation ->run() functions should use this function to expire
3111 * OpenFlow flows. */
3113 ofproto_rule_expire(struct rule *rule, uint8_t reason)
3115 struct ofproto *ofproto = rule->ofproto;
3116 struct ofopgroup *group;
3118 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
3120 ofproto_rule_send_removed(rule, reason);
3122 group = ofopgroup_create_unattached(ofproto);
3123 ofoperation_create(group, rule, OFOPERATION_DELETE);
3124 oftable_remove_rule(rule);
3125 ofproto->ofproto_class->rule_destruct(rule);
3126 ofopgroup_submit(group);
3130 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3132 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3133 struct ofputil_flow_mod fm;
3134 uint64_t ofpacts_stub[1024 / 8];
3135 struct ofpbuf ofpacts;
3139 error = reject_slave_controller(ofconn);
3144 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3145 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
3148 goto exit_free_ofpacts;
3151 /* We do not support the OpenFlow 1.0 emergency flow cache, which is not
3152 * required in OpenFlow 1.0.1 and removed from OpenFlow 1.1. */
3153 if (fm.flags & OFPFF_EMERG) {
3154 /* We do not support the emergency flow cache. It will hopefully get
3155 * dropped from OpenFlow in the near future. There is no good error
3156 * code, so just state that the flow table is full. */
3157 error = OFPERR_OFPFMFC_ALL_TABLES_FULL;
3159 error = handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
3162 goto exit_free_ofpacts;
3165 /* Record the operation for logging a summary report. */
3166 switch (fm.command) {
3172 case OFPFC_MODIFY_STRICT:
3173 ofproto->n_modify++;
3177 case OFPFC_DELETE_STRICT:
3178 ofproto->n_delete++;
3183 if (ofproto->next_op_report == LLONG_MAX) {
3184 ofproto->first_op = now;
3185 ofproto->next_op_report = MAX(now + 10 * 1000,
3186 ofproto->op_backoff);
3187 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
3189 ofproto->last_op = now;
3192 ofpbuf_uninit(&ofpacts);
3198 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
3199 const struct ofputil_flow_mod *fm,
3200 const struct ofp_header *oh)
3202 if (ofproto->n_pending >= 50) {
3203 assert(!list_is_empty(&ofproto->pending));
3204 return OFPROTO_POSTPONE;
3207 switch (fm->command) {
3209 return add_flow(ofproto, ofconn, fm, oh);
3212 return modify_flows_loose(ofproto, ofconn, fm, oh);
3214 case OFPFC_MODIFY_STRICT:
3215 return modify_flow_strict(ofproto, ofconn, fm, oh);
3218 return delete_flows_loose(ofproto, ofconn, fm, oh);
3220 case OFPFC_DELETE_STRICT:
3221 return delete_flow_strict(ofproto, ofconn, fm, oh);
3224 if (fm->command > 0xff) {
3225 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
3226 "flow_mod_table_id extension is not enabled",
3229 return OFPERR_OFPFMFC_BAD_COMMAND;
3234 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
3236 struct nx_role_request *nrr = (struct nx_role_request *) oh;
3237 struct nx_role_request *reply;
3241 role = ntohl(nrr->role);
3242 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
3243 && role != NX_ROLE_SLAVE) {
3244 return OFPERR_OFPRRFC_BAD_ROLE;
3247 if (ofconn_get_role(ofconn) != role
3248 && ofconn_has_pending_opgroups(ofconn)) {
3249 return OFPROTO_POSTPONE;
3252 ofconn_set_role(ofconn, role);
3254 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
3255 reply->role = htonl(role);
3256 ofconn_send_reply(ofconn, buf);
3262 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
3263 const struct ofp_header *oh)
3265 const struct nx_flow_mod_table_id *msg
3266 = (const struct nx_flow_mod_table_id *) oh;
3267 enum ofputil_protocol cur, next;
3269 cur = ofconn_get_protocol(ofconn);
3270 next = ofputil_protocol_set_tid(cur, msg->set != 0);
3271 ofconn_set_protocol(ofconn, next);
3277 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
3279 const struct nx_set_flow_format *msg
3280 = (const struct nx_set_flow_format *) oh;
3281 enum ofputil_protocol cur, next;
3282 enum ofputil_protocol next_base;
3284 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
3286 return OFPERR_OFPBRC_EPERM;
3289 cur = ofconn_get_protocol(ofconn);
3290 next = ofputil_protocol_set_base(cur, next_base);
3291 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
3292 /* Avoid sending async messages in surprising protocol. */
3293 return OFPROTO_POSTPONE;
3296 ofconn_set_protocol(ofconn, next);
3301 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
3302 const struct ofp_header *oh)
3304 const struct nx_set_packet_in_format *msg;
3307 msg = (const struct nx_set_packet_in_format *) oh;
3308 format = ntohl(msg->format);
3309 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
3310 return OFPERR_OFPBRC_EPERM;
3313 if (format != ofconn_get_packet_in_format(ofconn)
3314 && ofconn_has_pending_opgroups(ofconn)) {
3315 /* Avoid sending async message in surprsing packet in format. */
3316 return OFPROTO_POSTPONE;
3319 ofconn_set_packet_in_format(ofconn, format);
3324 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
3326 const struct nx_async_config *msg = (const struct nx_async_config *) oh;
3327 uint32_t master[OAM_N_TYPES];
3328 uint32_t slave[OAM_N_TYPES];
3330 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
3331 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
3332 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
3334 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
3335 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
3336 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
3338 ofconn_set_async_config(ofconn, master, slave);
3339 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
3340 !ofconn_get_miss_send_len(ofconn)) {
3341 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
3348 handle_nxt_set_controller_id(struct ofconn *ofconn,
3349 const struct ofp_header *oh)
3351 const struct nx_controller_id *nci;
3353 nci = (const struct nx_controller_id *) oh;
3354 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
3355 return OFPERR_NXBRC_MUST_BE_ZERO;
3358 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
3363 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
3367 if (ofconn_has_pending_opgroups(ofconn)) {
3368 return OFPROTO_POSTPONE;
3371 make_openflow_xid(sizeof *oh, OFPT10_BARRIER_REPLY, oh->xid, &buf);
3372 ofconn_send_reply(ofconn, buf);
3377 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
3379 const struct ofp_header *oh = msg->data;
3380 const struct ofputil_msg_type *type;
3383 error = ofputil_decode_msg_type(oh, &type);
3388 switch (ofputil_msg_type_code(type)) {
3389 /* OpenFlow requests. */
3390 case OFPUTIL_OFPT_ECHO_REQUEST:
3391 return handle_echo_request(ofconn, oh);
3393 case OFPUTIL_OFPT_FEATURES_REQUEST:
3394 return handle_features_request(ofconn, oh);
3396 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
3397 return handle_get_config_request(ofconn, oh);
3399 case OFPUTIL_OFPT_SET_CONFIG:
3400 return handle_set_config(ofconn, msg->data);
3402 case OFPUTIL_OFPT_PACKET_OUT:
3403 return handle_packet_out(ofconn, msg->data);
3405 case OFPUTIL_OFPT_PORT_MOD:
3406 return handle_port_mod(ofconn, oh);
3408 case OFPUTIL_OFPT_FLOW_MOD:
3409 return handle_flow_mod(ofconn, oh);
3411 case OFPUTIL_OFPT_BARRIER_REQUEST:
3412 return handle_barrier_request(ofconn, oh);
3414 /* OpenFlow replies. */
3415 case OFPUTIL_OFPT_ECHO_REPLY:
3418 /* Nicira extension requests. */
3419 case OFPUTIL_NXT_ROLE_REQUEST:
3420 return handle_role_request(ofconn, oh);
3422 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
3423 return handle_nxt_flow_mod_table_id(ofconn, oh);
3425 case OFPUTIL_NXT_SET_FLOW_FORMAT:
3426 return handle_nxt_set_flow_format(ofconn, oh);
3428 case OFPUTIL_NXT_SET_PACKET_IN_FORMAT:
3429 return handle_nxt_set_packet_in_format(ofconn, oh);
3431 case OFPUTIL_NXT_SET_CONTROLLER_ID:
3432 return handle_nxt_set_controller_id(ofconn, oh);
3434 case OFPUTIL_NXT_FLOW_MOD:
3435 return handle_flow_mod(ofconn, oh);
3437 case OFPUTIL_NXT_FLOW_AGE:
3438 /* Nothing to do. */
3441 case OFPUTIL_NXT_SET_ASYNC_CONFIG:
3442 return handle_nxt_set_async_config(ofconn, oh);
3444 /* Statistics requests. */
3445 case OFPUTIL_OFPST_DESC_REQUEST:
3446 return handle_desc_stats_request(ofconn, msg->data);
3448 case OFPUTIL_OFPST_FLOW_REQUEST:
3449 case OFPUTIL_NXST_FLOW_REQUEST:
3450 return handle_flow_stats_request(ofconn, msg->data);
3452 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
3453 case OFPUTIL_NXST_AGGREGATE_REQUEST:
3454 return handle_aggregate_stats_request(ofconn, msg->data);
3456 case OFPUTIL_OFPST_TABLE_REQUEST:
3457 return handle_table_stats_request(ofconn, msg->data);
3459 case OFPUTIL_OFPST_PORT_REQUEST:
3460 return handle_port_stats_request(ofconn, msg->data);
3462 case OFPUTIL_OFPST_QUEUE_REQUEST:
3463 return handle_queue_stats_request(ofconn, msg->data);
3465 case OFPUTIL_OFPST_PORT_DESC_REQUEST:
3466 return handle_port_desc_stats_request(ofconn, msg->data);
3468 case OFPUTIL_MSG_INVALID:
3469 case OFPUTIL_OFPT_HELLO:
3470 case OFPUTIL_OFPT_ERROR:
3471 case OFPUTIL_OFPT_FEATURES_REPLY:
3472 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
3473 case OFPUTIL_OFPT_PACKET_IN:
3474 case OFPUTIL_OFPT_FLOW_REMOVED:
3475 case OFPUTIL_OFPT_PORT_STATUS:
3476 case OFPUTIL_OFPT_BARRIER_REPLY:
3477 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
3478 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
3479 case OFPUTIL_OFPST_DESC_REPLY:
3480 case OFPUTIL_OFPST_FLOW_REPLY:
3481 case OFPUTIL_OFPST_QUEUE_REPLY:
3482 case OFPUTIL_OFPST_PORT_REPLY:
3483 case OFPUTIL_OFPST_TABLE_REPLY:
3484 case OFPUTIL_OFPST_AGGREGATE_REPLY:
3485 case OFPUTIL_OFPST_PORT_DESC_REPLY:
3486 case OFPUTIL_NXT_ROLE_REPLY:
3487 case OFPUTIL_NXT_FLOW_REMOVED:
3488 case OFPUTIL_NXT_PACKET_IN:
3489 case OFPUTIL_NXST_FLOW_REPLY:
3490 case OFPUTIL_NXST_AGGREGATE_REPLY:
3492 return (oh->type == OFPT10_STATS_REQUEST ||
3493 oh->type == OFPT10_STATS_REPLY
3494 ? OFPERR_OFPBRC_BAD_STAT
3495 : OFPERR_OFPBRC_BAD_TYPE);
3500 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
3502 int error = handle_openflow__(ofconn, ofp_msg);
3503 if (error && error != OFPROTO_POSTPONE) {
3504 ofconn_send_error(ofconn, ofp_msg->data, error);
3506 COVERAGE_INC(ofproto_recv_openflow);
3507 return error != OFPROTO_POSTPONE;
3510 /* Asynchronous operations. */
3512 /* Creates and returns a new ofopgroup that is not associated with any
3513 * OpenFlow connection.
3515 * The caller should add operations to the returned group with
3516 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3517 static struct ofopgroup *
3518 ofopgroup_create_unattached(struct ofproto *ofproto)
3520 struct ofopgroup *group = xzalloc(sizeof *group);
3521 group->ofproto = ofproto;
3522 list_init(&group->ofproto_node);
3523 list_init(&group->ops);
3524 list_init(&group->ofconn_node);
3528 /* Creates and returns a new ofopgroup for 'ofproto'.
3530 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3531 * connection. The 'request' and 'buffer_id' arguments are ignored.
3533 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3534 * If the ofopgroup eventually fails, then the error reply will include
3535 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3536 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3538 * The caller should add operations to the returned group with
3539 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3540 static struct ofopgroup *
3541 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3542 const struct ofp_header *request, uint32_t buffer_id)
3544 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3546 size_t request_len = ntohs(request->length);
3548 assert(ofconn_get_ofproto(ofconn) == ofproto);
3550 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3551 group->ofconn = ofconn;
3552 group->request = xmemdup(request, MIN(request_len, 64));
3553 group->buffer_id = buffer_id;
3558 /* Submits 'group' for processing.
3560 * If 'group' contains no operations (e.g. none were ever added, or all of the
3561 * ones that were added completed synchronously), then it is destroyed
3562 * immediately. Otherwise it is added to the ofproto's list of pending
3565 ofopgroup_submit(struct ofopgroup *group)
3567 if (list_is_empty(&group->ops)) {
3568 ofopgroup_destroy(group);
3570 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3571 group->ofproto->n_pending++;
3576 ofopgroup_destroy(struct ofopgroup *group)
3578 assert(list_is_empty(&group->ops));
3579 if (!list_is_empty(&group->ofproto_node)) {
3580 assert(group->ofproto->n_pending > 0);
3581 group->ofproto->n_pending--;
3582 list_remove(&group->ofproto_node);
3584 if (!list_is_empty(&group->ofconn_node)) {
3585 list_remove(&group->ofconn_node);
3587 ofconn_send_error(group->ofconn, group->request, group->error);
3589 connmgr_retry(group->ofproto->connmgr);
3591 free(group->request);
3595 /* Initiates a new operation on 'rule', of the specified 'type', within
3596 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3598 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3599 enum ofoperation_type type)
3601 struct ofproto *ofproto = group->ofproto;
3602 struct ofoperation *op;
3604 assert(!rule->pending);
3606 op = rule->pending = xzalloc(sizeof *op);
3608 list_push_back(&group->ops, &op->group_node);
3611 op->flow_cookie = rule->flow_cookie;
3613 if (type == OFOPERATION_DELETE) {
3614 hmap_insert(&ofproto->deletions, &op->hmap_node,
3615 cls_rule_hash(&rule->cr, rule->table_id));
3620 ofoperation_destroy(struct ofoperation *op)
3622 struct ofopgroup *group = op->group;
3625 op->rule->pending = NULL;
3627 if (op->type == OFOPERATION_DELETE) {
3628 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3630 list_remove(&op->group_node);
3634 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3635 ofopgroup_destroy(group);
3639 /* Indicates that 'op' completed with status 'error', which is either 0 to
3640 * indicate success or an OpenFlow error code on failure.
3642 * If 'error' is 0, indicating success, the operation will be committed
3643 * permanently to the flow table. There is one interesting subcase:
3645 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3646 * the flow table (the "victim" rule) by a new one, then the caller must
3647 * have uninitialized any derived state in the victim rule, as in step 5 in
3648 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3649 * performs steps 6 and 7 for the victim rule, most notably by calling its
3650 * ->rule_dealloc() function.
3652 * If 'error' is nonzero, then generally the operation will be rolled back:
3654 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3655 * restores the original rule. The caller must have uninitialized any
3656 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3657 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3658 * and 7 for the new rule, calling its ->rule_dealloc() function.
3660 * - If 'op' is a "modify flow" operation, ofproto restores the original
3663 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3664 * allowed to fail. It must always succeed.
3666 * Please see the large comment in ofproto/ofproto-provider.h titled
3667 * "Asynchronous Operation Support" for more information. */
3669 ofoperation_complete(struct ofoperation *op, enum ofperr error)
3671 struct ofopgroup *group = op->group;
3672 struct rule *rule = op->rule;
3673 struct ofproto *ofproto = rule->ofproto;
3675 assert(rule->pending == op);
3679 && op->type != OFOPERATION_DELETE
3681 && group->buffer_id != UINT32_MAX
3682 && list_is_singleton(&op->group_node)) {
3683 struct ofpbuf *packet;
3686 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3690 error = rule_execute(rule, in_port, packet);
3693 if (!group->error) {
3694 group->error = error;
3698 case OFOPERATION_ADD:
3700 ofproto_rule_destroy__(op->victim);
3701 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3702 == htons(VLAN_VID_MASK)) {
3703 if (ofproto->vlan_bitmap) {
3704 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3706 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3707 bitmap_set1(ofproto->vlan_bitmap, vid);
3708 ofproto->vlans_changed = true;
3711 ofproto->vlans_changed = true;
3715 oftable_substitute_rule(rule, op->victim);
3716 ofproto_rule_destroy__(rule);
3721 case OFOPERATION_DELETE:
3723 ofproto_rule_destroy__(rule);
3727 case OFOPERATION_MODIFY:
3729 rule->modified = time_msec();
3731 free(rule->ofpacts);
3732 rule->ofpacts = op->ofpacts;
3733 rule->ofpacts_len = op->ofpacts_len;
3741 ofoperation_destroy(op);
3745 ofoperation_get_victim(struct ofoperation *op)
3747 assert(op->type == OFOPERATION_ADD);
3752 pick_datapath_id(const struct ofproto *ofproto)
3754 const struct ofport *port;
3756 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3758 uint8_t ea[ETH_ADDR_LEN];
3761 error = netdev_get_etheraddr(port->netdev, ea);
3763 return eth_addr_to_uint64(ea);
3765 VLOG_WARN("%s: could not get MAC address for %s (%s)",
3766 ofproto->name, netdev_get_name(port->netdev),
3769 return ofproto->fallback_dpid;
3773 pick_fallback_dpid(void)
3775 uint8_t ea[ETH_ADDR_LEN];
3776 eth_addr_nicira_random(ea);
3777 return eth_addr_to_uint64(ea);
3780 /* Table overflow policy. */
3782 /* Chooses and returns a rule to evict from 'table'. Returns NULL if the table
3783 * is not configured to evict rules or if the table contains no evictable
3784 * rules. (Rules with 'evictable' set to false or with no timeouts are not
3786 static struct rule *
3787 choose_rule_to_evict(struct oftable *table)
3789 struct eviction_group *evg;
3791 if (!table->eviction_fields) {
3795 /* In the common case, the outer and inner loops here will each be entered
3798 * - The inner loop normally "return"s in its first iteration. If the
3799 * eviction group has any evictable rules, then it always returns in
3802 * - The outer loop only iterates more than once if the largest eviction
3803 * group has no evictable rules.
3805 * - The outer loop can exit only if table's 'max_flows' is all filled up
3806 * by unevictable rules'. */
3807 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
3810 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
3811 if (rule->evictable) {
3820 /* Searches 'ofproto' for tables that have more flows than their configured
3821 * maximum and that have flow eviction enabled, and evicts as many flows as
3822 * necessary and currently feasible from them.
3824 * This triggers only when an OpenFlow table has N flows in it and then the
3825 * client configures a maximum number of flows less than N. */
3827 ofproto_evict(struct ofproto *ofproto)
3829 struct ofopgroup *group;
3830 struct oftable *table;
3832 group = ofopgroup_create_unattached(ofproto);
3833 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
3834 while (classifier_count(&table->cls) > table->max_flows
3835 && table->eviction_fields) {
3838 rule = choose_rule_to_evict(table);
3839 if (!rule || rule->pending) {
3843 ofoperation_create(group, rule, OFOPERATION_DELETE);
3844 oftable_remove_rule(rule);
3845 ofproto->ofproto_class->rule_destruct(rule);
3848 ofopgroup_submit(group);
3851 /* Eviction groups. */
3853 /* Returns the priority to use for an eviction_group that contains 'n_rules'
3854 * rules. The priority contains low-order random bits to ensure that eviction
3855 * groups with the same number of rules are prioritized randomly. */
3857 eviction_group_priority(size_t n_rules)
3859 uint16_t size = MIN(UINT16_MAX, n_rules);
3860 return (size << 16) | random_uint16();
3863 /* Updates 'evg', an eviction_group within 'table', following a change that
3864 * adds or removes rules in 'evg'. */
3866 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
3868 heap_change(&table->eviction_groups_by_size, &evg->size_node,
3869 eviction_group_priority(heap_count(&evg->rules)));
3872 /* Destroys 'evg', an eviction_group within 'table':
3874 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
3875 * rules themselves, just removes them from the eviction group.)
3877 * - Removes 'evg' from 'table'.
3881 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
3883 while (!heap_is_empty(&evg->rules)) {
3886 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
3887 rule->eviction_group = NULL;
3889 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
3890 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
3891 heap_destroy(&evg->rules);
3895 /* Removes 'rule' from its eviction group, if any. */
3897 eviction_group_remove_rule(struct rule *rule)
3899 if (rule->eviction_group) {
3900 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3901 struct eviction_group *evg = rule->eviction_group;
3903 rule->eviction_group = NULL;
3904 heap_remove(&evg->rules, &rule->evg_node);
3905 if (heap_is_empty(&evg->rules)) {
3906 eviction_group_destroy(table, evg);
3908 eviction_group_resized(table, evg);
3913 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
3914 * returns the hash value. */
3916 eviction_group_hash_rule(struct rule *rule)
3918 struct oftable *table = &rule->ofproto->tables[rule->table_id];
3919 const struct mf_subfield *sf;
3922 hash = table->eviction_group_id_basis;
3923 for (sf = table->eviction_fields;
3924 sf < &table->eviction_fields[table->n_eviction_fields];
3927 if (mf_are_prereqs_ok(sf->field, &rule->cr.flow)) {
3928 union mf_value value;
3930 mf_get_value(sf->field, &rule->cr.flow, &value);
3932 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
3934 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
3935 unsigned int start = sf->ofs + sf->n_bits;
3936 bitwise_zero(&value, sf->field->n_bytes, start,
3937 sf->field->n_bytes * 8 - start);
3939 hash = hash_bytes(&value, sf->field->n_bytes, hash);
3941 hash = hash_int(hash, 0);
3948 /* Returns an eviction group within 'table' with the given 'id', creating one
3950 static struct eviction_group *
3951 eviction_group_find(struct oftable *table, uint32_t id)
3953 struct eviction_group *evg;
3955 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
3959 evg = xmalloc(sizeof *evg);
3960 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
3961 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
3962 eviction_group_priority(0));
3963 heap_init(&evg->rules);
3968 /* Returns an eviction priority for 'rule'. The return value should be
3969 * interpreted so that higher priorities make a rule more attractive candidates
3972 rule_eviction_priority(struct rule *rule)
3974 long long int hard_expiration;
3975 long long int idle_expiration;
3976 long long int expiration;
3977 uint32_t expiration_offset;
3979 /* Calculate time of expiration. */
3980 hard_expiration = (rule->hard_timeout
3981 ? rule->modified + rule->hard_timeout * 1000
3983 idle_expiration = (rule->idle_timeout
3984 ? rule->used + rule->idle_timeout * 1000
3986 expiration = MIN(hard_expiration, idle_expiration);
3987 if (expiration == LLONG_MAX) {
3991 /* Calculate the time of expiration as a number of (approximate) seconds
3992 * after program startup.
3994 * This should work OK for program runs that last UINT32_MAX seconds or
3995 * less. Therefore, please restart OVS at least once every 136 years. */
3996 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
3998 /* Invert the expiration offset because we're using a max-heap. */
3999 return UINT32_MAX - expiration_offset;
4002 /* Adds 'rule' to an appropriate eviction group for its oftable's
4003 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
4004 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
4007 * The caller must ensure that 'rule' is not already in an eviction group. */
4009 eviction_group_add_rule(struct rule *rule)
4011 struct ofproto *ofproto = rule->ofproto;
4012 struct oftable *table = &ofproto->tables[rule->table_id];
4014 if (table->eviction_fields
4015 && (rule->hard_timeout || rule->idle_timeout)) {
4016 struct eviction_group *evg;
4018 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
4020 rule->eviction_group = evg;
4021 heap_insert(&evg->rules, &rule->evg_node,
4022 rule_eviction_priority(rule));
4023 eviction_group_resized(table, evg);
4029 /* Initializes 'table'. */
4031 oftable_init(struct oftable *table)
4033 memset(table, 0, sizeof *table);
4034 classifier_init(&table->cls);
4035 table->max_flows = UINT_MAX;
4038 /* Destroys 'table', including its classifier and eviction groups.
4040 * The caller is responsible for freeing 'table' itself. */
4042 oftable_destroy(struct oftable *table)
4044 assert(classifier_is_empty(&table->cls));
4045 oftable_disable_eviction(table);
4046 classifier_destroy(&table->cls);
4050 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
4051 * string, then 'table' will use its default name.
4053 * This only affects the name exposed for a table exposed through the OpenFlow
4054 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
4056 oftable_set_name(struct oftable *table, const char *name)
4058 if (name && name[0]) {
4059 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
4060 if (!table->name || strncmp(name, table->name, len)) {
4062 table->name = xmemdup0(name, len);
4070 /* oftables support a choice of two policies when adding a rule would cause the
4071 * number of flows in the table to exceed the configured maximum number: either
4072 * they can refuse to add the new flow or they can evict some existing flow.
4073 * This function configures the former policy on 'table'. */
4075 oftable_disable_eviction(struct oftable *table)
4077 if (table->eviction_fields) {
4078 struct eviction_group *evg, *next;
4080 HMAP_FOR_EACH_SAFE (evg, next, id_node,
4081 &table->eviction_groups_by_id) {
4082 eviction_group_destroy(table, evg);
4084 hmap_destroy(&table->eviction_groups_by_id);
4085 heap_destroy(&table->eviction_groups_by_size);
4087 free(table->eviction_fields);
4088 table->eviction_fields = NULL;
4089 table->n_eviction_fields = 0;
4093 /* oftables support a choice of two policies when adding a rule would cause the
4094 * number of flows in the table to exceed the configured maximum number: either
4095 * they can refuse to add the new flow or they can evict some existing flow.
4096 * This function configures the latter policy on 'table', with fairness based
4097 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
4098 * 'n_fields' as 0 disables fairness.) */
4100 oftable_enable_eviction(struct oftable *table,
4101 const struct mf_subfield *fields, size_t n_fields)
4103 struct cls_cursor cursor;
4106 if (table->eviction_fields
4107 && n_fields == table->n_eviction_fields
4109 || !memcmp(fields, table->eviction_fields,
4110 n_fields * sizeof *fields))) {
4115 oftable_disable_eviction(table);
4117 table->n_eviction_fields = n_fields;
4118 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
4120 table->eviction_group_id_basis = random_uint32();
4121 hmap_init(&table->eviction_groups_by_id);
4122 heap_init(&table->eviction_groups_by_size);
4124 cls_cursor_init(&cursor, &table->cls, NULL);
4125 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4126 eviction_group_add_rule(rule);
4130 /* Removes 'rule' from the oftable that contains it. */
4132 oftable_remove_rule(struct rule *rule)
4134 struct ofproto *ofproto = rule->ofproto;
4135 struct oftable *table = &ofproto->tables[rule->table_id];
4137 classifier_remove(&table->cls, &rule->cr);
4138 eviction_group_remove_rule(rule);
4141 /* Inserts 'rule' into its oftable. Removes any existing rule from 'rule''s
4142 * oftable that has an identical cls_rule. Returns the rule that was removed,
4143 * if any, and otherwise NULL. */
4144 static struct rule *
4145 oftable_replace_rule(struct rule *rule)
4147 struct ofproto *ofproto = rule->ofproto;
4148 struct oftable *table = &ofproto->tables[rule->table_id];
4149 struct rule *victim;
4151 victim = rule_from_cls_rule(classifier_replace(&table->cls, &rule->cr));
4153 eviction_group_remove_rule(victim);
4155 eviction_group_add_rule(rule);
4159 /* Removes 'old' from its oftable then, if 'new' is nonnull, inserts 'new'. */
4161 oftable_substitute_rule(struct rule *old, struct rule *new)
4164 oftable_replace_rule(new);
4166 oftable_remove_rule(old);
4170 /* unixctl commands. */
4173 ofproto_lookup(const char *name)
4175 struct ofproto *ofproto;
4177 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
4179 if (!strcmp(ofproto->name, name)) {
4187 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
4188 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
4190 struct ofproto *ofproto;
4194 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
4195 ds_put_format(&results, "%s\n", ofproto->name);
4197 unixctl_command_reply(conn, ds_cstr(&results));
4198 ds_destroy(&results);
4202 ofproto_unixctl_init(void)
4204 static bool registered;
4210 unixctl_command_register("ofproto/list", "", 0, 0,
4211 ofproto_unixctl_list, NULL);
4214 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4216 * This is deprecated. It is only for compatibility with broken device drivers
4217 * in old versions of Linux that do not properly support VLANs when VLAN
4218 * devices are not used. When broken device drivers are no longer in
4219 * widespread use, we will delete these interfaces. */
4221 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
4222 * (exactly) by an OpenFlow rule in 'ofproto'. */
4224 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
4226 const struct oftable *oftable;
4228 free(ofproto->vlan_bitmap);
4229 ofproto->vlan_bitmap = bitmap_allocate(4096);
4230 ofproto->vlans_changed = false;
4232 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
4233 const struct cls_table *table;
4235 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
4236 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
4237 == htons(VLAN_VID_MASK)) {
4238 const struct cls_rule *rule;
4240 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
4241 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
4242 bitmap_set1(vlan_bitmap, vid);
4243 bitmap_set1(ofproto->vlan_bitmap, vid);
4250 /* Returns true if new VLANs have come into use by the flow table since the
4251 * last call to ofproto_get_vlan_usage().
4253 * We don't track when old VLANs stop being used. */
4255 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
4257 return ofproto->vlans_changed;
4260 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
4261 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
4262 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
4263 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
4264 * then the VLAN device is un-enslaved. */
4266 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
4267 uint16_t realdev_ofp_port, int vid)
4269 struct ofport *ofport;
4272 assert(vlandev_ofp_port != realdev_ofp_port);
4274 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
4276 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
4277 ofproto->name, vlandev_ofp_port);
4281 if (!ofproto->ofproto_class->set_realdev) {
4282 if (!vlandev_ofp_port) {
4285 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
4289 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
4291 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
4292 ofproto->name, vlandev_ofp_port,
4293 netdev_get_name(ofport->netdev), strerror(error));