2 * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
25 #include "byte-order.h"
26 #include "classifier.h"
29 #include "dynamic-string.h"
34 #include "ofp-print.h"
37 #include "ofproto-provider.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
43 #include "poll-loop.h"
47 #include "unaligned.h"
51 VLOG_DEFINE_THIS_MODULE(ofproto);
53 COVERAGE_DEFINE(ofproto_error);
54 COVERAGE_DEFINE(ofproto_flush);
55 COVERAGE_DEFINE(ofproto_no_packet_in);
56 COVERAGE_DEFINE(ofproto_packet_out);
57 COVERAGE_DEFINE(ofproto_queue_req);
58 COVERAGE_DEFINE(ofproto_recv_openflow);
59 COVERAGE_DEFINE(ofproto_reinit_ports);
60 COVERAGE_DEFINE(ofproto_uninstallable);
61 COVERAGE_DEFINE(ofproto_update_port);
64 S_OPENFLOW, /* Processing OpenFlow commands. */
65 S_FLUSH, /* Deleting all flow table rules. */
68 enum ofoperation_type {
74 /* A single OpenFlow request can execute any number of operations. The
75 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
76 * ofconn to which an error reply should be sent if necessary.
78 * ofproto initiates some operations internally. These operations are still
79 * assigned to groups but will not have an associated ofconn. */
81 struct ofproto *ofproto; /* Owning ofproto. */
82 struct list ofproto_node; /* In ofproto's "pending" list. */
83 struct list ops; /* List of "struct ofoperation"s. */
85 /* Data needed to send OpenFlow reply on failure or to send a buffered
88 * If list_is_empty(ofconn_node) then this ofopgroup never had an
89 * associated ofconn or its ofconn's connection dropped after it initiated
90 * the operation. In the latter case 'ofconn' is a wild pointer that
91 * refers to freed memory, so the 'ofconn' member must be used only if
92 * !list_is_empty(ofconn_node).
94 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
95 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
96 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
97 uint32_t buffer_id; /* Buffer id from original request. */
98 int error; /* 0 if no error yet, otherwise error code. */
101 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
102 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
103 const struct ofp_header *,
105 static void ofopgroup_submit(struct ofopgroup *);
106 static void ofopgroup_destroy(struct ofopgroup *);
108 /* A single flow table operation. */
110 struct ofopgroup *group; /* Owning group. */
111 struct list group_node; /* In ofopgroup's "ops" list. */
112 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
113 struct rule *rule; /* Rule being operated upon. */
114 enum ofoperation_type type; /* Type of operation. */
115 int status; /* -1 if pending, otherwise 0 or error code. */
116 struct rule *victim; /* OFOPERATION_ADDING: Replaced rule. */
117 union ofp_action *actions; /* OFOPERATION_MODIFYING: Replaced actions. */
118 int n_actions; /* OFOPERATION_MODIFYING: # of old actions. */
119 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
122 static void ofoperation_create(struct ofopgroup *, struct rule *,
123 enum ofoperation_type);
124 static void ofoperation_destroy(struct ofoperation *);
126 static void ofport_destroy__(struct ofport *);
127 static void ofport_destroy(struct ofport *);
129 static uint64_t pick_datapath_id(const struct ofproto *);
130 static uint64_t pick_fallback_dpid(void);
132 static void ofproto_destroy__(struct ofproto *);
134 static void ofproto_rule_destroy__(struct rule *);
135 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
137 static void ofopgroup_destroy(struct ofopgroup *);
139 static int add_flow(struct ofproto *, struct ofconn *,
140 const struct ofputil_flow_mod *,
141 const struct ofp_header *);
143 static bool handle_openflow(struct ofconn *, struct ofpbuf *);
144 static int handle_flow_mod__(struct ofproto *, struct ofconn *,
145 const struct ofputil_flow_mod *,
146 const struct ofp_header *);
148 static void update_port(struct ofproto *, const char *devname);
149 static int init_ports(struct ofproto *);
150 static void reinit_ports(struct ofproto *);
151 static void set_internal_devs_mtu(struct ofproto *);
153 static void ofproto_unixctl_init(void);
155 /* All registered ofproto classes, in probe order. */
156 static const struct ofproto_class **ofproto_classes;
157 static size_t n_ofproto_classes;
158 static size_t allocated_ofproto_classes;
160 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
161 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
166 ofproto_initialize(void)
172 ofproto_class_register(&ofproto_dpif_class);
176 /* 'type' should be a normalized datapath type, as returned by
177 * ofproto_normalize_type(). Returns the corresponding ofproto_class
178 * structure, or a null pointer if there is none registered for 'type'. */
179 static const struct ofproto_class *
180 ofproto_class_find__(const char *type)
184 ofproto_initialize();
185 for (i = 0; i < n_ofproto_classes; i++) {
186 const struct ofproto_class *class = ofproto_classes[i];
191 class->enumerate_types(&types);
192 found = sset_contains(&types, type);
193 sset_destroy(&types);
199 VLOG_WARN("unknown datapath type %s", type);
203 /* Registers a new ofproto class. After successful registration, new ofprotos
204 * of that type can be created using ofproto_create(). */
206 ofproto_class_register(const struct ofproto_class *new_class)
210 for (i = 0; i < n_ofproto_classes; i++) {
211 if (ofproto_classes[i] == new_class) {
216 if (n_ofproto_classes >= allocated_ofproto_classes) {
217 ofproto_classes = x2nrealloc(ofproto_classes,
218 &allocated_ofproto_classes,
219 sizeof *ofproto_classes);
221 ofproto_classes[n_ofproto_classes++] = new_class;
225 /* Unregisters a datapath provider. 'type' must have been previously
226 * registered and not currently be in use by any ofprotos. After
227 * unregistration new datapaths of that type cannot be opened using
228 * ofproto_create(). */
230 ofproto_class_unregister(const struct ofproto_class *class)
234 for (i = 0; i < n_ofproto_classes; i++) {
235 if (ofproto_classes[i] == class) {
236 for (i++; i < n_ofproto_classes; i++) {
237 ofproto_classes[i - 1] = ofproto_classes[i];
243 VLOG_WARN("attempted to unregister an ofproto class that is not "
248 /* Clears 'types' and enumerates all registered ofproto types into it. The
249 * caller must first initialize the sset. */
251 ofproto_enumerate_types(struct sset *types)
255 ofproto_initialize();
256 for (i = 0; i < n_ofproto_classes; i++) {
257 ofproto_classes[i]->enumerate_types(types);
261 /* Returns the fully spelled out name for the given ofproto 'type'.
263 * Normalized type string can be compared with strcmp(). Unnormalized type
264 * string might be the same even if they have different spellings. */
266 ofproto_normalize_type(const char *type)
268 return type && type[0] ? type : "system";
271 /* Clears 'names' and enumerates the names of all known created ofprotos with
272 * the given 'type'. The caller must first initialize the sset. Returns 0 if
273 * successful, otherwise a positive errno value.
275 * Some kinds of datapaths might not be practically enumerable. This is not
276 * considered an error. */
278 ofproto_enumerate_names(const char *type, struct sset *names)
280 const struct ofproto_class *class = ofproto_class_find__(type);
281 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
285 ofproto_create(const char *datapath_name, const char *datapath_type,
286 struct ofproto **ofprotop)
288 const struct ofproto_class *class;
289 struct classifier *table;
290 struct ofproto *ofproto;
296 ofproto_initialize();
297 ofproto_unixctl_init();
299 datapath_type = ofproto_normalize_type(datapath_type);
300 class = ofproto_class_find__(datapath_type);
302 VLOG_WARN("could not create datapath %s of unknown type %s",
303 datapath_name, datapath_type);
307 ofproto = class->alloc();
309 VLOG_ERR("failed to allocate datapath %s of type %s",
310 datapath_name, datapath_type);
315 memset(ofproto, 0, sizeof *ofproto);
316 ofproto->ofproto_class = class;
317 ofproto->name = xstrdup(datapath_name);
318 ofproto->type = xstrdup(datapath_type);
319 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
320 hash_string(ofproto->name, 0));
321 ofproto->datapath_id = 0;
322 ofproto_set_flow_eviction_threshold(ofproto,
323 OFPROTO_FLOW_EVICTON_THRESHOLD_DEFAULT);
324 ofproto->forward_bpdu = false;
325 ofproto->fallback_dpid = pick_fallback_dpid();
326 ofproto->mfr_desc = xstrdup(DEFAULT_MFR_DESC);
327 ofproto->hw_desc = xstrdup(DEFAULT_HW_DESC);
328 ofproto->sw_desc = xstrdup(DEFAULT_SW_DESC);
329 ofproto->serial_desc = xstrdup(DEFAULT_SERIAL_DESC);
330 ofproto->dp_desc = xstrdup(DEFAULT_DP_DESC);
331 ofproto->frag_handling = OFPC_FRAG_NORMAL;
332 hmap_init(&ofproto->ports);
333 shash_init(&ofproto->port_by_name);
334 ofproto->tables = NULL;
335 ofproto->n_tables = 0;
336 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
337 ofproto->state = S_OPENFLOW;
338 list_init(&ofproto->pending);
339 ofproto->n_pending = 0;
340 hmap_init(&ofproto->deletions);
341 ofproto->vlan_bitmap = NULL;
342 ofproto->vlans_changed = false;
344 error = ofproto->ofproto_class->construct(ofproto, &n_tables);
346 VLOG_ERR("failed to open datapath %s: %s",
347 datapath_name, strerror(error));
348 ofproto_destroy__(ofproto);
352 assert(n_tables >= 1 && n_tables <= 255);
353 ofproto->n_tables = n_tables;
354 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
355 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
356 classifier_init(table);
359 ofproto->datapath_id = pick_datapath_id(ofproto);
360 VLOG_INFO("using datapath ID %016"PRIx64, ofproto->datapath_id);
368 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
370 uint64_t old_dpid = p->datapath_id;
371 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
372 if (p->datapath_id != old_dpid) {
373 VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id);
375 /* Force all active connections to reconnect, since there is no way to
376 * notify a controller that the datapath ID has changed. */
377 ofproto_reconnect_controllers(p);
382 ofproto_set_controllers(struct ofproto *p,
383 const struct ofproto_controller *controllers,
384 size_t n_controllers)
386 connmgr_set_controllers(p->connmgr, controllers, n_controllers);
390 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
392 connmgr_set_fail_mode(p->connmgr, fail_mode);
395 /* Drops the connections between 'ofproto' and all of its controllers, forcing
396 * them to reconnect. */
398 ofproto_reconnect_controllers(struct ofproto *ofproto)
400 connmgr_reconnect(ofproto->connmgr);
403 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
404 * in-band control should guarantee access, in the same way that in-band
405 * control guarantees access to OpenFlow controllers. */
407 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
408 const struct sockaddr_in *extras, size_t n)
410 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
413 /* Sets the OpenFlow queue used by flows set up by in-band control on
414 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
415 * flows will use the default queue. */
417 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
419 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
422 /* Sets the number of flows at which eviction from the kernel flow table
425 ofproto_set_flow_eviction_threshold(struct ofproto *ofproto, unsigned threshold)
427 if (threshold < OFPROTO_FLOW_EVICTION_THRESHOLD_MIN) {
428 ofproto->flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_MIN;
430 ofproto->flow_eviction_threshold = threshold;
434 /* If forward_bpdu is true, the NORMAL action will forward frames with
435 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
436 * the NORMAL action will drop these frames. */
438 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
440 bool old_val = ofproto->forward_bpdu;
441 ofproto->forward_bpdu = forward_bpdu;
442 if (old_val != ofproto->forward_bpdu) {
443 if (ofproto->ofproto_class->forward_bpdu_changed) {
444 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
449 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
450 * 'idle_time', in seconds. */
452 ofproto_set_mac_idle_time(struct ofproto *ofproto, unsigned idle_time)
454 if (ofproto->ofproto_class->set_mac_idle_time) {
455 ofproto->ofproto_class->set_mac_idle_time(ofproto, idle_time);
460 ofproto_set_desc(struct ofproto *p,
461 const char *mfr_desc, const char *hw_desc,
462 const char *sw_desc, const char *serial_desc,
465 struct ofp_desc_stats *ods;
468 if (strlen(mfr_desc) >= sizeof ods->mfr_desc) {
469 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
470 sizeof ods->mfr_desc);
473 p->mfr_desc = xstrdup(mfr_desc);
476 if (strlen(hw_desc) >= sizeof ods->hw_desc) {
477 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
478 sizeof ods->hw_desc);
481 p->hw_desc = xstrdup(hw_desc);
484 if (strlen(sw_desc) >= sizeof ods->sw_desc) {
485 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
486 sizeof ods->sw_desc);
489 p->sw_desc = xstrdup(sw_desc);
492 if (strlen(serial_desc) >= sizeof ods->serial_num) {
493 VLOG_WARN("truncating serial_desc, must be less than %zu "
495 sizeof ods->serial_num);
497 free(p->serial_desc);
498 p->serial_desc = xstrdup(serial_desc);
501 if (strlen(dp_desc) >= sizeof ods->dp_desc) {
502 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
503 sizeof ods->dp_desc);
506 p->dp_desc = xstrdup(dp_desc);
511 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
513 return connmgr_set_snoops(ofproto->connmgr, snoops);
517 ofproto_set_netflow(struct ofproto *ofproto,
518 const struct netflow_options *nf_options)
520 if (nf_options && sset_is_empty(&nf_options->collectors)) {
524 if (ofproto->ofproto_class->set_netflow) {
525 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
527 return nf_options ? EOPNOTSUPP : 0;
532 ofproto_set_sflow(struct ofproto *ofproto,
533 const struct ofproto_sflow_options *oso)
535 if (oso && sset_is_empty(&oso->targets)) {
539 if (ofproto->ofproto_class->set_sflow) {
540 return ofproto->ofproto_class->set_sflow(ofproto, oso);
542 return oso ? EOPNOTSUPP : 0;
546 /* Spanning Tree Protocol (STP) configuration. */
548 /* Configures STP on 'ofproto' using the settings defined in 's'. If
549 * 's' is NULL, disables STP.
551 * Returns 0 if successful, otherwise a positive errno value. */
553 ofproto_set_stp(struct ofproto *ofproto,
554 const struct ofproto_stp_settings *s)
556 return (ofproto->ofproto_class->set_stp
557 ? ofproto->ofproto_class->set_stp(ofproto, s)
561 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
562 * 'enabled' member of 's' is false, then the other members are not
565 * Returns 0 if successful, otherwise a positive errno value. */
567 ofproto_get_stp_status(struct ofproto *ofproto,
568 struct ofproto_stp_status *s)
570 return (ofproto->ofproto_class->get_stp_status
571 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
575 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
576 * in 's'. The caller is responsible for assigning STP port numbers
577 * (using the 'port_num' member in the range of 1 through 255, inclusive)
578 * and ensuring there are no duplicates. If the 's' is NULL, then STP
579 * is disabled on the port.
581 * Returns 0 if successful, otherwise a positive errno value.*/
583 ofproto_port_set_stp(struct ofproto *ofproto, uint16_t ofp_port,
584 const struct ofproto_port_stp_settings *s)
586 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
588 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
589 ofproto->name, ofp_port);
593 return (ofproto->ofproto_class->set_stp_port
594 ? ofproto->ofproto_class->set_stp_port(ofport, s)
598 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
599 * 's'. If the 'enabled' member in 's' is false, then the other members
600 * are not meaningful.
602 * Returns 0 if successful, otherwise a positive errno value.*/
604 ofproto_port_get_stp_status(struct ofproto *ofproto, uint16_t ofp_port,
605 struct ofproto_port_stp_status *s)
607 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
609 VLOG_WARN("%s: cannot get STP status on nonexistent port %"PRIu16,
610 ofproto->name, ofp_port);
614 return (ofproto->ofproto_class->get_stp_port_status
615 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
619 /* Queue DSCP configuration. */
621 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
622 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
623 * to implement QoS. Instead, it is used to implement features which require
624 * knowledge of what queues exist on a port, and some basic information about
627 * Returns 0 if successful, otherwise a positive errno value. */
629 ofproto_port_set_queues(struct ofproto *ofproto, uint16_t ofp_port,
630 const struct ofproto_port_queue *queues,
633 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
636 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
637 ofproto->name, ofp_port);
641 return (ofproto->ofproto_class->set_queues
642 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
646 /* Connectivity Fault Management configuration. */
648 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
650 ofproto_port_clear_cfm(struct ofproto *ofproto, uint16_t ofp_port)
652 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
653 if (ofport && ofproto->ofproto_class->set_cfm) {
654 ofproto->ofproto_class->set_cfm(ofport, NULL);
658 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
659 * basic configuration from the configuration members in 'cfm', and the remote
660 * maintenance point ID from remote_mpid. Ignores the statistics members of
663 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
665 ofproto_port_set_cfm(struct ofproto *ofproto, uint16_t ofp_port,
666 const struct cfm_settings *s)
668 struct ofport *ofport;
671 ofport = ofproto_get_port(ofproto, ofp_port);
673 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
674 ofproto->name, ofp_port);
678 /* XXX: For configuration simplicity, we only support one remote_mpid
679 * outside of the CFM module. It's not clear if this is the correct long
680 * term solution or not. */
681 error = (ofproto->ofproto_class->set_cfm
682 ? ofproto->ofproto_class->set_cfm(ofport, s)
685 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
686 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
691 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
692 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
693 * 0 if LACP partner information is not current (generally indicating a
694 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
696 ofproto_port_is_lacp_current(struct ofproto *ofproto, uint16_t ofp_port)
698 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
699 return (ofport && ofproto->ofproto_class->port_is_lacp_current
700 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
706 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
707 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
708 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
709 * configuration plus, if there is more than one slave, a bonding
712 * If 'aux' is already registered then this function updates its configuration
713 * to 's'. Otherwise, this function registers a new bundle.
715 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
718 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
719 const struct ofproto_bundle_settings *s)
721 return (ofproto->ofproto_class->bundle_set
722 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
726 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
727 * If no such bundle has been registered, this has no effect. */
729 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
731 return ofproto_bundle_register(ofproto, aux, NULL);
735 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
736 * If 'aux' is already registered then this function updates its configuration
737 * to 's'. Otherwise, this function registers a new mirror. */
739 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
740 const struct ofproto_mirror_settings *s)
742 return (ofproto->ofproto_class->mirror_set
743 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
747 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
748 * If no mirror has been registered, this has no effect. */
750 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
752 return ofproto_mirror_register(ofproto, aux, NULL);
755 /* Retrieves statistics from mirror associated with client data pointer
756 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
757 * 'bytes', respectively. If a particular counters is not supported,
758 * the appropriate argument is set to UINT64_MAX. */
760 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
761 uint64_t *packets, uint64_t *bytes)
763 if (!ofproto->ofproto_class->mirror_get_stats) {
764 *packets = *bytes = UINT64_MAX;
768 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
772 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
773 * which all packets are flooded, instead of using MAC learning. If
774 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
776 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
779 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
781 return (ofproto->ofproto_class->set_flood_vlans
782 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
786 /* Returns true if 'aux' is a registered bundle that is currently in use as the
787 * output for a mirror. */
789 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
791 return (ofproto->ofproto_class->is_mirror_output_bundle
792 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
797 ofproto_has_snoops(const struct ofproto *ofproto)
799 return connmgr_has_snoops(ofproto->connmgr);
803 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
805 connmgr_get_snoops(ofproto->connmgr, snoops);
809 ofproto_flush__(struct ofproto *ofproto)
811 struct classifier *table;
812 struct ofopgroup *group;
814 if (ofproto->ofproto_class->flush) {
815 ofproto->ofproto_class->flush(ofproto);
818 group = ofopgroup_create_unattached(ofproto);
819 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
820 struct rule *rule, *next_rule;
821 struct cls_cursor cursor;
823 cls_cursor_init(&cursor, table, NULL);
824 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
825 if (!rule->pending) {
826 ofoperation_create(group, rule, OFOPERATION_DELETE);
827 classifier_remove(table, &rule->cr);
828 ofproto->ofproto_class->rule_destruct(rule);
832 ofopgroup_submit(group);
836 ofproto_destroy__(struct ofproto *ofproto)
838 struct classifier *table;
840 assert(list_is_empty(&ofproto->pending));
841 assert(!ofproto->n_pending);
843 connmgr_destroy(ofproto->connmgr);
845 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
848 free(ofproto->mfr_desc);
849 free(ofproto->hw_desc);
850 free(ofproto->sw_desc);
851 free(ofproto->serial_desc);
852 free(ofproto->dp_desc);
853 hmap_destroy(&ofproto->ports);
854 shash_destroy(&ofproto->port_by_name);
856 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
857 assert(classifier_is_empty(table));
858 classifier_destroy(table);
860 free(ofproto->tables);
862 hmap_destroy(&ofproto->deletions);
864 free(ofproto->vlan_bitmap);
866 ofproto->ofproto_class->dealloc(ofproto);
870 ofproto_destroy(struct ofproto *p)
872 struct ofport *ofport, *next_ofport;
879 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
880 ofport_destroy(ofport);
883 p->ofproto_class->destruct(p);
884 ofproto_destroy__(p);
887 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
888 * kernel datapath, for example, this destroys the datapath in the kernel, and
889 * with the netdev-based datapath, it tears down the data structures that
890 * represent the datapath.
892 * The datapath should not be currently open as an ofproto. */
894 ofproto_delete(const char *name, const char *type)
896 const struct ofproto_class *class = ofproto_class_find__(type);
897 return (!class ? EAFNOSUPPORT
898 : !class->del ? EACCES
899 : class->del(type, name));
903 process_port_change(struct ofproto *ofproto, int error, char *devname)
905 if (error == ENOBUFS) {
906 reinit_ports(ofproto);
908 update_port(ofproto, devname);
914 ofproto_run(struct ofproto *p)
916 struct sset changed_netdevs;
917 const char *changed_netdev;
918 struct ofport *ofport;
921 error = p->ofproto_class->run(p);
922 if (error && error != EAGAIN) {
923 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, strerror(error));
926 if (p->ofproto_class->port_poll) {
929 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
930 process_port_change(p, error, devname);
934 /* Update OpenFlow port status for any port whose netdev has changed.
936 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
937 * destroyed, so it's not safe to update ports directly from the
938 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
939 * need this two-phase approach. */
940 sset_init(&changed_netdevs);
941 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
942 unsigned int change_seq = netdev_change_seq(ofport->netdev);
943 if (ofport->change_seq != change_seq) {
944 ofport->change_seq = change_seq;
945 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
948 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
949 update_port(p, changed_netdev);
951 sset_destroy(&changed_netdevs);
956 connmgr_run(p->connmgr, handle_openflow);
960 connmgr_run(p->connmgr, NULL);
962 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
963 connmgr_flushed(p->connmgr);
964 p->state = S_OPENFLOW;
975 /* Performs periodic activity required by 'ofproto' that needs to be done
976 * with the least possible latency.
978 * It makes sense to call this function a couple of times per poll loop, to
979 * provide a significant performance boost on some benchmarks with the
980 * ofproto-dpif implementation. */
982 ofproto_run_fast(struct ofproto *p)
986 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
987 if (error && error != EAGAIN) {
988 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
989 p->name, strerror(error));
995 ofproto_wait(struct ofproto *p)
997 struct ofport *ofport;
999 p->ofproto_class->wait(p);
1000 if (p->ofproto_class->port_poll_wait) {
1001 p->ofproto_class->port_poll_wait(p);
1004 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1005 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1006 poll_immediate_wake();
1012 connmgr_wait(p->connmgr, true);
1016 connmgr_wait(p->connmgr, false);
1017 if (list_is_empty(&p->pending) && hmap_is_empty(&p->deletions)) {
1018 poll_immediate_wake();
1025 ofproto_is_alive(const struct ofproto *p)
1027 return connmgr_has_controllers(p->connmgr);
1031 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1034 connmgr_get_controller_info(ofproto->connmgr, info);
1038 ofproto_free_ofproto_controller_info(struct shash *info)
1040 connmgr_free_controller_info(info);
1043 /* Makes a deep copy of 'old' into 'port'. */
1045 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1047 port->name = xstrdup(old->name);
1048 port->type = xstrdup(old->type);
1049 port->ofp_port = old->ofp_port;
1052 /* Frees memory allocated to members of 'ofproto_port'.
1054 * Do not call this function on an ofproto_port obtained from
1055 * ofproto_port_dump_next(): that function retains ownership of the data in the
1058 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1060 free(ofproto_port->name);
1061 free(ofproto_port->type);
1064 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1066 * This function provides no status indication. An error status for the entire
1067 * dump operation is provided when it is completed by calling
1068 * ofproto_port_dump_done().
1071 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1072 const struct ofproto *ofproto)
1074 dump->ofproto = ofproto;
1075 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1079 /* Attempts to retrieve another port from 'dump', which must have been created
1080 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1081 * 'port' and returns true. On failure, returns false.
1083 * Failure might indicate an actual error or merely that the last port has been
1084 * dumped. An error status for the entire dump operation is provided when it
1085 * is completed by calling ofproto_port_dump_done().
1087 * The ofproto owns the data stored in 'port'. It will remain valid until at
1088 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1089 * ofproto_port_dump_done(). */
1091 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1092 struct ofproto_port *port)
1094 const struct ofproto *ofproto = dump->ofproto;
1100 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1103 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1109 /* Completes port table dump operation 'dump', which must have been created
1110 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1111 * error-free, otherwise a positive errno value describing the problem. */
1113 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1115 const struct ofproto *ofproto = dump->ofproto;
1117 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1120 return dump->error == EOF ? 0 : dump->error;
1123 /* Attempts to add 'netdev' as a port on 'ofproto'. If successful, returns 0
1124 * and sets '*ofp_portp' to the new port's OpenFlow port number (if 'ofp_portp'
1125 * is non-null). On failure, returns a positive errno value and sets
1126 * '*ofp_portp' to OFPP_NONE (if 'ofp_portp' is non-null). */
1128 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1129 uint16_t *ofp_portp)
1134 error = ofproto->ofproto_class->port_add(ofproto, netdev, &ofp_port);
1136 update_port(ofproto, netdev_get_name(netdev));
1139 *ofp_portp = error ? OFPP_NONE : ofp_port;
1144 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1145 * initializes '*port' appropriately; on failure, returns a positive errno
1148 * The caller owns the data in 'ofproto_port' and must free it with
1149 * ofproto_port_destroy() when it is no longer needed. */
1151 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1152 struct ofproto_port *port)
1156 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1158 memset(port, 0, sizeof *port);
1163 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1164 * Returns 0 if successful, otherwise a positive errno. */
1166 ofproto_port_del(struct ofproto *ofproto, uint16_t ofp_port)
1168 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1169 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1172 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1173 if (!error && ofport) {
1174 /* 'name' is the netdev's name and update_port() is going to close the
1175 * netdev. Just in case update_port() refers to 'name' after it
1176 * destroys 'ofport', make a copy of it around the update_port()
1178 char *devname = xstrdup(name);
1179 update_port(ofproto, devname);
1185 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1186 * performs the 'n_actions' actions in 'actions'. The new flow will not
1189 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1190 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1191 * controllers; otherwise, it will be hidden.
1193 * The caller retains ownership of 'cls_rule' and 'actions'.
1195 * This is a helper function for in-band control and fail-open. */
1197 ofproto_add_flow(struct ofproto *ofproto, const struct cls_rule *cls_rule,
1198 const union ofp_action *actions, size_t n_actions)
1200 const struct rule *rule;
1202 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1203 &ofproto->tables[0], cls_rule));
1204 if (!rule || !ofputil_actions_equal(rule->actions, rule->n_actions,
1205 actions, n_actions)) {
1206 struct ofputil_flow_mod fm;
1208 memset(&fm, 0, sizeof fm);
1210 fm.buffer_id = UINT32_MAX;
1211 fm.actions = (union ofp_action *) actions;
1212 fm.n_actions = n_actions;
1213 add_flow(ofproto, NULL, &fm, NULL);
1217 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1218 * OpenFlow error code as encoded by ofp_mkerr() on failure, or
1219 * OFPROTO_POSTPONE if the operation cannot be initiated now but may be retried
1222 * This is a helper function for in-band control and fail-open. */
1224 ofproto_flow_mod(struct ofproto *ofproto, const struct ofputil_flow_mod *fm)
1226 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1229 /* Searches for a rule with matching criteria exactly equal to 'target' in
1230 * ofproto's table 0 and, if it finds one, deletes it.
1232 * This is a helper function for in-band control and fail-open. */
1234 ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target)
1238 rule = rule_from_cls_rule(classifier_find_rule_exactly(
1239 &ofproto->tables[0], target));
1241 /* No such rule -> success. */
1243 } else if (rule->pending) {
1244 /* An operation on the rule is already pending -> failure.
1245 * Caller must retry later if it's important. */
1248 /* Initiate deletion -> success. */
1249 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
1250 ofoperation_create(group, rule, OFOPERATION_DELETE);
1251 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
1252 rule->ofproto->ofproto_class->rule_destruct(rule);
1253 ofopgroup_submit(group);
1259 /* Starts the process of deleting all of the flows from all of ofproto's flow
1260 * tables and then reintroducing the flows required by in-band control and
1261 * fail-open. The process will complete in a later call to ofproto_run(). */
1263 ofproto_flush_flows(struct ofproto *ofproto)
1265 COVERAGE_INC(ofproto_flush);
1266 ofproto->state = S_FLUSH;
1270 reinit_ports(struct ofproto *p)
1272 struct ofproto_port_dump dump;
1273 struct sset devnames;
1274 struct ofport *ofport;
1275 struct ofproto_port ofproto_port;
1276 const char *devname;
1278 COVERAGE_INC(ofproto_reinit_ports);
1280 sset_init(&devnames);
1281 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1282 sset_add(&devnames, netdev_get_name(ofport->netdev));
1284 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1285 sset_add(&devnames, ofproto_port.name);
1288 SSET_FOR_EACH (devname, &devnames) {
1289 update_port(p, devname);
1291 sset_destroy(&devnames);
1294 /* Opens and returns a netdev for 'ofproto_port', or a null pointer if the
1295 * netdev cannot be opened. On success, also fills in 'opp'. */
1296 static struct netdev *
1297 ofport_open(const struct ofproto_port *ofproto_port, struct ofp_phy_port *opp)
1299 uint32_t curr, advertised, supported, peer;
1300 enum netdev_flags flags;
1301 struct netdev *netdev;
1304 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
1306 VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s "
1307 "cannot be opened (%s)",
1308 ofproto_port->name, ofproto_port->ofp_port,
1309 ofproto_port->name, strerror(error));
1313 netdev_get_flags(netdev, &flags);
1314 netdev_get_features(netdev, &curr, &advertised, &supported, &peer);
1316 opp->port_no = htons(ofproto_port->ofp_port);
1317 netdev_get_etheraddr(netdev, opp->hw_addr);
1318 ovs_strzcpy(opp->name, ofproto_port->name, sizeof opp->name);
1319 opp->config = flags & NETDEV_UP ? 0 : htonl(OFPPC_PORT_DOWN);
1320 opp->state = netdev_get_carrier(netdev) ? 0 : htonl(OFPPS_LINK_DOWN);
1321 opp->curr = htonl(curr);
1322 opp->advertised = htonl(advertised);
1323 opp->supported = htonl(supported);
1324 opp->peer = htonl(peer);
1329 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
1330 * port number, and 'config' bits other than OFPPC_PORT_DOWN are
1333 ofport_equal(const struct ofp_phy_port *a, const struct ofp_phy_port *b)
1335 BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */
1336 return (!memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr)
1337 && a->state == b->state
1338 && !((a->config ^ b->config) & htonl(OFPPC_PORT_DOWN))
1339 && a->curr == b->curr
1340 && a->advertised == b->advertised
1341 && a->supported == b->supported
1342 && a->peer == b->peer);
1345 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
1346 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
1347 * one with the same name or port number). */
1349 ofport_install(struct ofproto *p,
1350 struct netdev *netdev, const struct ofp_phy_port *opp)
1352 const char *netdev_name = netdev_get_name(netdev);
1353 struct ofport *ofport;
1357 /* Create ofport. */
1358 ofport = p->ofproto_class->port_alloc();
1363 ofport->ofproto = p;
1364 ofport->netdev = netdev;
1365 ofport->change_seq = netdev_change_seq(netdev);
1367 ofport->ofp_port = ntohs(opp->port_no);
1369 /* Add port to 'p'. */
1370 hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->ofp_port, 0));
1371 shash_add(&p->port_by_name, netdev_name, ofport);
1373 if (!netdev_get_mtu(netdev, &dev_mtu)) {
1374 ofport->mtu = dev_mtu;
1375 set_internal_devs_mtu(p);
1380 /* Let the ofproto_class initialize its private data. */
1381 error = p->ofproto_class->port_construct(ofport);
1385 connmgr_send_port_status(p->connmgr, opp, OFPPR_ADD);
1389 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
1390 p->name, netdev_name, strerror(error));
1392 ofport_destroy__(ofport);
1394 netdev_close(netdev);
1398 /* Removes 'ofport' from 'p' and destroys it. */
1400 ofport_remove(struct ofport *ofport)
1402 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->opp,
1404 ofport_destroy(ofport);
1407 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
1410 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
1412 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
1414 ofport_remove(port);
1418 /* Updates 'port' with new 'opp' description.
1420 * Does not handle a name or port number change. The caller must implement
1421 * such a change as a delete followed by an add. */
1423 ofport_modified(struct ofport *port, struct ofp_phy_port *opp)
1425 memcpy(port->opp.hw_addr, opp->hw_addr, ETH_ADDR_LEN);
1426 port->opp.config = ((port->opp.config & ~htonl(OFPPC_PORT_DOWN))
1427 | (opp->config & htonl(OFPPC_PORT_DOWN)));
1428 port->opp.state = opp->state;
1429 port->opp.curr = opp->curr;
1430 port->opp.advertised = opp->advertised;
1431 port->opp.supported = opp->supported;
1432 port->opp.peer = opp->peer;
1434 connmgr_send_port_status(port->ofproto->connmgr, &port->opp, OFPPR_MODIFY);
1437 /* Update OpenFlow 'state' in 'port' and notify controller. */
1439 ofproto_port_set_state(struct ofport *port, ovs_be32 state)
1441 if (port->opp.state != state) {
1442 port->opp.state = state;
1443 connmgr_send_port_status(port->ofproto->connmgr, &port->opp,
1449 ofproto_port_unregister(struct ofproto *ofproto, uint16_t ofp_port)
1451 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
1453 if (port->ofproto->ofproto_class->set_realdev) {
1454 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
1456 if (port->ofproto->ofproto_class->set_stp_port) {
1457 port->ofproto->ofproto_class->set_stp_port(port, NULL);
1459 if (port->ofproto->ofproto_class->set_cfm) {
1460 port->ofproto->ofproto_class->set_cfm(port, NULL);
1462 if (port->ofproto->ofproto_class->bundle_remove) {
1463 port->ofproto->ofproto_class->bundle_remove(port);
1469 ofport_destroy__(struct ofport *port)
1471 struct ofproto *ofproto = port->ofproto;
1472 const char *name = netdev_get_name(port->netdev);
1474 hmap_remove(&ofproto->ports, &port->hmap_node);
1475 shash_delete(&ofproto->port_by_name,
1476 shash_find(&ofproto->port_by_name, name));
1478 netdev_close(port->netdev);
1479 ofproto->ofproto_class->port_dealloc(port);
1483 ofport_destroy(struct ofport *port)
1486 port->ofproto->ofproto_class->port_destruct(port);
1487 ofport_destroy__(port);
1492 ofproto_get_port(const struct ofproto *ofproto, uint16_t ofp_port)
1494 struct ofport *port;
1496 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node,
1497 hash_int(ofp_port, 0), &ofproto->ports) {
1498 if (port->ofp_port == ofp_port) {
1506 update_port(struct ofproto *ofproto, const char *name)
1508 struct ofproto_port ofproto_port;
1509 struct ofp_phy_port opp;
1510 struct netdev *netdev;
1511 struct ofport *port;
1513 COVERAGE_INC(ofproto_update_port);
1515 /* Fetch 'name''s location and properties from the datapath. */
1516 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
1517 ? ofport_open(&ofproto_port, &opp)
1520 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
1521 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
1522 struct netdev *old_netdev = port->netdev;
1525 /* 'name' hasn't changed location. Any properties changed? */
1526 if (!ofport_equal(&port->opp, &opp)) {
1527 ofport_modified(port, &opp);
1530 /* If this is a non-internal port and the MTU changed, check
1531 * if the datapath's MTU needs to be updated. */
1532 if (strcmp(netdev_get_type(netdev), "internal")
1533 && !netdev_get_mtu(netdev, &dev_mtu)
1534 && port->mtu != dev_mtu) {
1535 set_internal_devs_mtu(ofproto);
1536 port->mtu = dev_mtu;
1539 /* Install the newly opened netdev in case it has changed.
1540 * Don't close the old netdev yet in case port_modified has to
1541 * remove a retained reference to it.*/
1542 port->netdev = netdev;
1543 port->change_seq = netdev_change_seq(netdev);
1545 if (port->ofproto->ofproto_class->port_modified) {
1546 port->ofproto->ofproto_class->port_modified(port);
1549 netdev_close(old_netdev);
1551 /* If 'port' is nonnull then its name differs from 'name' and thus
1552 * we should delete it. If we think there's a port named 'name'
1553 * then its port number must be wrong now so delete it too. */
1555 ofport_remove(port);
1557 ofport_remove_with_name(ofproto, name);
1558 ofport_install(ofproto, netdev, &opp);
1561 /* Any port named 'name' is gone now. */
1562 ofport_remove_with_name(ofproto, name);
1564 ofproto_port_destroy(&ofproto_port);
1568 init_ports(struct ofproto *p)
1570 struct ofproto_port_dump dump;
1571 struct ofproto_port ofproto_port;
1573 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1574 uint16_t ofp_port = ofproto_port.ofp_port;
1575 if (ofproto_get_port(p, ofp_port)) {
1576 VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath",
1578 } else if (shash_find(&p->port_by_name, ofproto_port.name)) {
1579 VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath",
1582 struct ofp_phy_port opp;
1583 struct netdev *netdev;
1585 netdev = ofport_open(&ofproto_port, &opp);
1587 ofport_install(p, netdev, &opp);
1595 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
1596 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
1598 find_min_mtu(struct ofproto *p)
1600 struct ofport *ofport;
1603 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1604 struct netdev *netdev = ofport->netdev;
1607 /* Skip any internal ports, since that's what we're trying to
1609 if (!strcmp(netdev_get_type(netdev), "internal")) {
1613 if (netdev_get_mtu(netdev, &dev_mtu)) {
1616 if (!mtu || dev_mtu < mtu) {
1621 return mtu ? mtu: ETH_PAYLOAD_MAX;
1624 /* Set the MTU of all datapath devices on 'p' to the minimum of the
1625 * non-datapath ports. */
1627 set_internal_devs_mtu(struct ofproto *p)
1629 struct ofport *ofport;
1630 int mtu = find_min_mtu(p);
1632 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1633 struct netdev *netdev = ofport->netdev;
1635 if (!strcmp(netdev_get_type(netdev), "internal")) {
1636 netdev_set_mtu(netdev, mtu);
1642 ofproto_rule_destroy__(struct rule *rule)
1644 free(rule->actions);
1645 rule->ofproto->ofproto_class->rule_dealloc(rule);
1648 /* This function allows an ofproto implementation to destroy any rules that
1649 * remain when its ->destruct() function is called. The caller must have
1650 * already uninitialized any derived members of 'rule' (step 5 described in the
1651 * large comment in ofproto/ofproto-provider.h titled "Life Cycle").
1652 * This function implements steps 6 and 7.
1654 * This function should only be called from an ofproto implementation's
1655 * ->destruct() function. It is not suitable elsewhere. */
1657 ofproto_rule_destroy(struct rule *rule)
1659 assert(!rule->pending);
1660 classifier_remove(&rule->ofproto->tables[rule->table_id], &rule->cr);
1661 ofproto_rule_destroy__(rule);
1664 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1665 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1668 rule_has_out_port(const struct rule *rule, uint16_t out_port)
1670 const union ofp_action *oa;
1673 if (out_port == OFPP_NONE) {
1676 OFPUTIL_ACTION_FOR_EACH_UNSAFE (oa, left, rule->actions, rule->n_actions) {
1677 if (action_outputs_to_port(oa, htons(out_port))) {
1684 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
1685 * statistics appropriately. 'packet' must have at least sizeof(struct
1686 * ofp_packet_in) bytes of headroom.
1688 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
1689 * with statistics for 'packet' either way.
1691 * Takes ownership of 'packet'. */
1693 rule_execute(struct rule *rule, uint16_t in_port, struct ofpbuf *packet)
1697 assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in));
1699 flow_extract(packet, 0, 0, in_port, &flow);
1700 return rule->ofproto->ofproto_class->rule_execute(rule, &flow, packet);
1703 /* Returns true if 'rule' should be hidden from the controller.
1705 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1706 * (e.g. by in-band control) and are intentionally hidden from the
1709 rule_is_hidden(const struct rule *rule)
1711 return rule->cr.priority > UINT16_MAX;
1715 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
1717 ofconn_send_reply(ofconn, make_echo_reply(oh));
1722 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
1724 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1725 struct ofp_switch_features *osf;
1727 struct ofport *port;
1731 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip, &actions);
1732 assert(actions & (1 << OFPAT_OUTPUT)); /* sanity check */
1734 osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf);
1735 osf->datapath_id = htonll(ofproto->datapath_id);
1736 osf->n_buffers = htonl(pktbuf_capacity());
1737 osf->n_tables = ofproto->n_tables;
1738 osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS |
1739 OFPC_PORT_STATS | OFPC_QUEUE_STATS);
1741 osf->capabilities |= htonl(OFPC_ARP_MATCH_IP);
1743 osf->actions = htonl(actions);
1745 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
1746 ofpbuf_put(buf, &port->opp, sizeof port->opp);
1749 ofconn_send_reply(ofconn, buf);
1754 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
1756 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1757 struct ofp_switch_config *osc;
1761 osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf);
1762 osc->flags = htons(ofproto->frag_handling);
1763 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
1764 ofconn_send_reply(ofconn, buf);
1770 handle_set_config(struct ofconn *ofconn, const struct ofp_switch_config *osc)
1772 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
1773 uint16_t flags = ntohs(osc->flags);
1775 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
1776 || ofconn_get_role(ofconn) != NX_ROLE_SLAVE) {
1777 enum ofp_config_flags cur = ofproto->frag_handling;
1778 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
1780 assert((cur & OFPC_FRAG_MASK) == cur);
1782 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
1783 ofproto->frag_handling = next;
1785 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
1787 ofputil_frag_handling_to_string(next));
1792 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
1797 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
1798 * error message code (composed with ofp_mkerr()) for the caller to propagate
1799 * upward. Otherwise, returns 0. */
1801 reject_slave_controller(const struct ofconn *ofconn)
1803 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
1804 && ofconn_get_role(ofconn) == NX_ROLE_SLAVE) {
1805 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
1812 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
1814 struct ofproto *p = ofconn_get_ofproto(ofconn);
1815 struct ofp_packet_out *opo;
1816 struct ofpbuf payload, *buffer;
1817 union ofp_action *ofp_actions;
1818 struct ofpbuf request;
1820 size_t n_ofp_actions;
1824 COVERAGE_INC(ofproto_packet_out);
1826 error = reject_slave_controller(ofconn);
1831 /* Get ofp_packet_out. */
1832 ofpbuf_use_const(&request, oh, ntohs(oh->length));
1833 opo = ofpbuf_pull(&request, offsetof(struct ofp_packet_out, actions));
1836 error = ofputil_pull_actions(&request, ntohs(opo->actions_len),
1837 &ofp_actions, &n_ofp_actions);
1843 if (opo->buffer_id != htonl(UINT32_MAX)) {
1844 error = ofconn_pktbuf_retrieve(ofconn, ntohl(opo->buffer_id),
1846 if (error || !buffer) {
1855 /* Get in_port and partially validate it.
1857 * We don't know what range of ports the ofproto actually implements, but
1858 * we do know that only certain reserved ports (numbered OFPP_MAX and
1859 * above) are valid. */
1860 in_port = ntohs(opo->in_port);
1861 if (in_port >= OFPP_MAX && in_port != OFPP_LOCAL && in_port != OFPP_NONE) {
1862 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT);
1865 /* Send out packet. */
1866 flow_extract(&payload, 0, 0, in_port, &flow);
1867 error = p->ofproto_class->packet_out(p, &payload, &flow,
1868 ofp_actions, n_ofp_actions);
1869 ofpbuf_delete(buffer);
1875 update_port_config(struct ofport *port, ovs_be32 config, ovs_be32 mask)
1877 ovs_be32 old_config = port->opp.config;
1879 mask &= config ^ port->opp.config;
1880 if (mask & htonl(OFPPC_PORT_DOWN)) {
1881 if (config & htonl(OFPPC_PORT_DOWN)) {
1882 netdev_turn_flags_off(port->netdev, NETDEV_UP, true);
1884 netdev_turn_flags_on(port->netdev, NETDEV_UP, true);
1888 port->opp.config ^= mask & (htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP |
1889 OFPPC_NO_FLOOD | OFPPC_NO_FWD |
1890 OFPPC_NO_PACKET_IN));
1891 if (port->opp.config != old_config) {
1892 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
1897 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
1899 struct ofproto *p = ofconn_get_ofproto(ofconn);
1900 const struct ofp_port_mod *opm = (const struct ofp_port_mod *) oh;
1901 struct ofport *port;
1904 error = reject_slave_controller(ofconn);
1909 port = ofproto_get_port(p, ntohs(opm->port_no));
1911 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT);
1912 } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) {
1913 return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR);
1915 update_port_config(port, opm->config, opm->mask);
1916 if (opm->advertise) {
1917 netdev_set_advertisements(port->netdev, ntohl(opm->advertise));
1924 handle_desc_stats_request(struct ofconn *ofconn,
1925 const struct ofp_stats_msg *request)
1927 struct ofproto *p = ofconn_get_ofproto(ofconn);
1928 struct ofp_desc_stats *ods;
1931 ods = ofputil_make_stats_reply(sizeof *ods, request, &msg);
1932 ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc);
1933 ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc);
1934 ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc);
1935 ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num);
1936 ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc);
1937 ofconn_send_reply(ofconn, msg);
1943 handle_table_stats_request(struct ofconn *ofconn,
1944 const struct ofp_stats_msg *request)
1946 struct ofproto *p = ofconn_get_ofproto(ofconn);
1947 struct ofp_table_stats *ots;
1951 ofputil_make_stats_reply(sizeof(struct ofp_stats_msg), request, &msg);
1953 ots = ofpbuf_put_zeros(msg, sizeof *ots * p->n_tables);
1954 for (i = 0; i < p->n_tables; i++) {
1955 ots[i].table_id = i;
1956 sprintf(ots[i].name, "table%zu", i);
1957 ots[i].wildcards = htonl(OFPFW_ALL);
1958 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
1959 ots[i].active_count = htonl(classifier_count(&p->tables[i]));
1962 p->ofproto_class->get_tables(p, ots);
1964 ofconn_send_reply(ofconn, msg);
1969 append_port_stat(struct ofport *port, struct list *replies)
1971 struct netdev_stats stats;
1972 struct ofp_port_stats *ops;
1974 /* Intentionally ignore return value, since errors will set
1975 * 'stats' to all-1s, which is correct for OpenFlow, and
1976 * netdev_get_stats() will log errors. */
1977 netdev_get_stats(port->netdev, &stats);
1979 ops = ofputil_append_stats_reply(sizeof *ops, replies);
1980 ops->port_no = port->opp.port_no;
1981 memset(ops->pad, 0, sizeof ops->pad);
1982 put_32aligned_be64(&ops->rx_packets, htonll(stats.rx_packets));
1983 put_32aligned_be64(&ops->tx_packets, htonll(stats.tx_packets));
1984 put_32aligned_be64(&ops->rx_bytes, htonll(stats.rx_bytes));
1985 put_32aligned_be64(&ops->tx_bytes, htonll(stats.tx_bytes));
1986 put_32aligned_be64(&ops->rx_dropped, htonll(stats.rx_dropped));
1987 put_32aligned_be64(&ops->tx_dropped, htonll(stats.tx_dropped));
1988 put_32aligned_be64(&ops->rx_errors, htonll(stats.rx_errors));
1989 put_32aligned_be64(&ops->tx_errors, htonll(stats.tx_errors));
1990 put_32aligned_be64(&ops->rx_frame_err, htonll(stats.rx_frame_errors));
1991 put_32aligned_be64(&ops->rx_over_err, htonll(stats.rx_over_errors));
1992 put_32aligned_be64(&ops->rx_crc_err, htonll(stats.rx_crc_errors));
1993 put_32aligned_be64(&ops->collisions, htonll(stats.collisions));
1997 handle_port_stats_request(struct ofconn *ofconn,
1998 const struct ofp_port_stats_request *psr)
2000 struct ofproto *p = ofconn_get_ofproto(ofconn);
2001 struct ofport *port;
2002 struct list replies;
2004 ofputil_start_stats_reply(&psr->osm, &replies);
2005 if (psr->port_no != htons(OFPP_NONE)) {
2006 port = ofproto_get_port(p, ntohs(psr->port_no));
2008 append_port_stat(port, &replies);
2011 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
2012 append_port_stat(port, &replies);
2016 ofconn_send_replies(ofconn, &replies);
2021 calc_flow_duration__(long long int start, uint32_t *sec, uint32_t *nsec)
2023 long long int msecs = time_msec() - start;
2024 *sec = msecs / 1000;
2025 *nsec = (msecs % 1000) * (1000 * 1000);
2028 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
2029 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
2031 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
2033 return (table_id == 0xff || table_id < ofproto->n_tables
2035 : ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_TABLE_ID));
2039 static struct classifier *
2040 first_matching_table(struct ofproto *ofproto, uint8_t table_id)
2042 if (table_id == 0xff) {
2043 return &ofproto->tables[0];
2044 } else if (table_id < ofproto->n_tables) {
2045 return &ofproto->tables[table_id];
2051 static struct classifier *
2052 next_matching_table(struct ofproto *ofproto,
2053 struct classifier *cls, uint8_t table_id)
2055 return (table_id == 0xff && cls != &ofproto->tables[ofproto->n_tables - 1]
2060 /* Assigns CLS to each classifier table, in turn, that matches TABLE_ID in
2063 * - If TABLE_ID is 0xff, this iterates over every classifier table in
2066 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
2067 * only once, for that table.
2069 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
2070 * entered at all. (Perhaps you should have validated TABLE_ID with
2071 * check_table_id().)
2073 * All parameters are evaluated multiple times.
2075 #define FOR_EACH_MATCHING_TABLE(CLS, TABLE_ID, OFPROTO) \
2076 for ((CLS) = first_matching_table(OFPROTO, TABLE_ID); \
2078 (CLS) = next_matching_table(OFPROTO, CLS, TABLE_ID))
2080 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2081 * 'table_id' is 0xff) that match 'match' in the "loose" way required for
2082 * OpenFlow OFPFC_MODIFY and OFPFC_DELETE requests and puts them on list
2085 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2086 * to 'out_port' are included.
2088 * Hidden rules are always omitted.
2090 * Returns 0 on success, otherwise an OpenFlow error code. */
2092 collect_rules_loose(struct ofproto *ofproto, uint8_t table_id,
2093 const struct cls_rule *match, uint16_t out_port,
2096 struct classifier *cls;
2099 error = check_table_id(ofproto, table_id);
2105 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
2106 struct cls_cursor cursor;
2109 cls_cursor_init(&cursor, cls, match);
2110 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2111 if (rule->pending) {
2112 return OFPROTO_POSTPONE;
2114 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
2115 list_push_back(rules, &rule->ofproto_node);
2122 /* Searches 'ofproto' for rules in table 'table_id' (or in all tables, if
2123 * 'table_id' is 0xff) that match 'match' in the "strict" way required for
2124 * OpenFlow OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests and puts them
2127 * If 'out_port' is anything other than OFPP_NONE, then only rules that output
2128 * to 'out_port' are included.
2130 * Hidden rules are always omitted.
2132 * Returns 0 on success, otherwise an OpenFlow error code. */
2134 collect_rules_strict(struct ofproto *ofproto, uint8_t table_id,
2135 const struct cls_rule *match, uint16_t out_port,
2138 struct classifier *cls;
2141 error = check_table_id(ofproto, table_id);
2147 FOR_EACH_MATCHING_TABLE (cls, table_id, ofproto) {
2150 rule = rule_from_cls_rule(classifier_find_rule_exactly(cls, match));
2152 if (rule->pending) {
2153 return OFPROTO_POSTPONE;
2155 if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) {
2156 list_push_back(rules, &rule->ofproto_node);
2164 handle_flow_stats_request(struct ofconn *ofconn,
2165 const struct ofp_stats_msg *osm)
2167 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2168 struct ofputil_flow_stats_request fsr;
2169 struct list replies;
2174 error = ofputil_decode_flow_stats_request(&fsr, &osm->header);
2179 error = collect_rules_loose(ofproto, fsr.table_id, &fsr.match,
2180 fsr.out_port, &rules);
2185 ofputil_start_stats_reply(osm, &replies);
2186 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2187 struct ofputil_flow_stats fs;
2190 fs.cookie = rule->flow_cookie;
2191 fs.table_id = rule->table_id;
2192 calc_flow_duration__(rule->created, &fs.duration_sec,
2194 fs.idle_timeout = rule->idle_timeout;
2195 fs.hard_timeout = rule->hard_timeout;
2196 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
2198 fs.actions = rule->actions;
2199 fs.n_actions = rule->n_actions;
2200 ofputil_append_flow_stats_reply(&fs, &replies);
2202 ofconn_send_replies(ofconn, &replies);
2208 flow_stats_ds(struct rule *rule, struct ds *results)
2210 uint64_t packet_count, byte_count;
2212 rule->ofproto->ofproto_class->rule_get_stats(rule,
2213 &packet_count, &byte_count);
2215 if (rule->table_id != 0) {
2216 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
2218 ds_put_format(results, "duration=%llds, ",
2219 (time_msec() - rule->created) / 1000);
2220 ds_put_format(results, "priority=%u, ", rule->cr.priority);
2221 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
2222 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
2223 cls_rule_format(&rule->cr, results);
2224 ds_put_char(results, ',');
2225 if (rule->n_actions > 0) {
2226 ofp_print_actions(results, rule->actions, rule->n_actions);
2228 ds_put_cstr(results, "drop");
2230 ds_put_cstr(results, "\n");
2233 /* Adds a pretty-printed description of all flows to 'results', including
2234 * hidden flows (e.g., set up by in-band control). */
2236 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
2238 struct classifier *cls;
2240 OFPROTO_FOR_EACH_TABLE (cls, p) {
2241 struct cls_cursor cursor;
2244 cls_cursor_init(&cursor, cls, NULL);
2245 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
2246 flow_stats_ds(rule, results);
2251 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
2252 * '*engine_type' and '*engine_id', respectively. */
2254 ofproto_get_netflow_ids(const struct ofproto *ofproto,
2255 uint8_t *engine_type, uint8_t *engine_id)
2257 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
2260 /* Checks the fault status of CFM for 'ofp_port' within 'ofproto'. Returns 1
2261 * if CFM is faulted (generally indiciating a connectivity problem), 0 if CFM
2262 * is not faulted, and -1 if CFM is not enabled on 'ofp_port'. */
2264 ofproto_port_get_cfm_fault(const struct ofproto *ofproto, uint16_t ofp_port)
2266 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2267 return (ofport && ofproto->ofproto_class->get_cfm_fault
2268 ? ofproto->ofproto_class->get_cfm_fault(ofport)
2272 /* Gets the MPIDs of the remote maintenance points broadcasting to 'ofp_port'
2273 * within 'ofproto'. Populates 'rmps' with an array of MPIDs owned by
2274 * 'ofproto', and 'n_rmps' with the number of MPIDs in 'rmps'. Returns a
2275 * number less than 0 if CFM is not enabled on 'ofp_port'. */
2277 ofproto_port_get_cfm_remote_mpids(const struct ofproto *ofproto,
2278 uint16_t ofp_port, const uint64_t **rmps,
2281 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2285 return (ofport && ofproto->ofproto_class->get_cfm_remote_mpids
2286 ? ofproto->ofproto_class->get_cfm_remote_mpids(ofport, rmps,
2292 handle_aggregate_stats_request(struct ofconn *ofconn,
2293 const struct ofp_stats_msg *osm)
2295 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2296 struct ofputil_flow_stats_request request;
2297 struct ofputil_aggregate_stats stats;
2298 bool unknown_packets, unknown_bytes;
2299 struct ofpbuf *reply;
2304 error = ofputil_decode_flow_stats_request(&request, &osm->header);
2309 error = collect_rules_loose(ofproto, request.table_id, &request.match,
2310 request.out_port, &rules);
2315 memset(&stats, 0, sizeof stats);
2316 unknown_packets = unknown_bytes = false;
2317 LIST_FOR_EACH (rule, ofproto_node, &rules) {
2318 uint64_t packet_count;
2319 uint64_t byte_count;
2321 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
2324 if (packet_count == UINT64_MAX) {
2325 unknown_packets = true;
2327 stats.packet_count += packet_count;
2330 if (byte_count == UINT64_MAX) {
2331 unknown_bytes = true;
2333 stats.byte_count += byte_count;
2338 if (unknown_packets) {
2339 stats.packet_count = UINT64_MAX;
2341 if (unknown_bytes) {
2342 stats.byte_count = UINT64_MAX;
2345 reply = ofputil_encode_aggregate_stats_reply(&stats, osm);
2346 ofconn_send_reply(ofconn, reply);
2351 struct queue_stats_cbdata {
2352 struct ofport *ofport;
2353 struct list replies;
2357 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
2358 const struct netdev_queue_stats *stats)
2360 struct ofp_queue_stats *reply;
2362 reply = ofputil_append_stats_reply(sizeof *reply, &cbdata->replies);
2363 reply->port_no = cbdata->ofport->opp.port_no;
2364 memset(reply->pad, 0, sizeof reply->pad);
2365 reply->queue_id = htonl(queue_id);
2366 put_32aligned_be64(&reply->tx_bytes, htonll(stats->tx_bytes));
2367 put_32aligned_be64(&reply->tx_packets, htonll(stats->tx_packets));
2368 put_32aligned_be64(&reply->tx_errors, htonll(stats->tx_errors));
2372 handle_queue_stats_dump_cb(uint32_t queue_id,
2373 struct netdev_queue_stats *stats,
2376 struct queue_stats_cbdata *cbdata = cbdata_;
2378 put_queue_stats(cbdata, queue_id, stats);
2382 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
2383 struct queue_stats_cbdata *cbdata)
2385 cbdata->ofport = port;
2386 if (queue_id == OFPQ_ALL) {
2387 netdev_dump_queue_stats(port->netdev,
2388 handle_queue_stats_dump_cb, cbdata);
2390 struct netdev_queue_stats stats;
2392 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
2393 put_queue_stats(cbdata, queue_id, &stats);
2399 handle_queue_stats_request(struct ofconn *ofconn,
2400 const struct ofp_queue_stats_request *qsr)
2402 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2403 struct queue_stats_cbdata cbdata;
2404 struct ofport *port;
2405 unsigned int port_no;
2408 COVERAGE_INC(ofproto_queue_req);
2410 ofputil_start_stats_reply(&qsr->osm, &cbdata.replies);
2412 port_no = ntohs(qsr->port_no);
2413 queue_id = ntohl(qsr->queue_id);
2414 if (port_no == OFPP_ALL) {
2415 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2416 handle_queue_stats_for_port(port, queue_id, &cbdata);
2418 } else if (port_no < OFPP_MAX) {
2419 port = ofproto_get_port(ofproto, port_no);
2421 handle_queue_stats_for_port(port, queue_id, &cbdata);
2424 ofpbuf_list_delete(&cbdata.replies);
2425 return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT);
2427 ofconn_send_replies(ofconn, &cbdata.replies);
2433 is_flow_deletion_pending(const struct ofproto *ofproto,
2434 const struct cls_rule *cls_rule,
2437 if (!hmap_is_empty(&ofproto->deletions)) {
2438 struct ofoperation *op;
2440 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
2441 cls_rule_hash(cls_rule, table_id),
2442 &ofproto->deletions) {
2443 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
2452 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
2453 * in which no matching flow already exists in the flow table.
2455 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
2456 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
2457 * error code as encoded by ofp_mkerr() on failure, or OFPROTO_POSTPONE if the
2458 * operation cannot be initiated now but may be retried later.
2460 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
2463 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
2464 const struct ofputil_flow_mod *fm, const struct ofp_header *request)
2466 struct classifier *table;
2467 struct ofopgroup *group;
2468 struct rule *victim;
2472 error = check_table_id(ofproto, fm->table_id);
2478 if (fm->table_id == 0xff) {
2480 if (ofproto->ofproto_class->rule_choose_table) {
2481 error = ofproto->ofproto_class->rule_choose_table(ofproto, &fm->cr,
2486 assert(table_id < ofproto->n_tables);
2487 table = &ofproto->tables[table_id];
2489 table = &ofproto->tables[0];
2491 } else if (fm->table_id < ofproto->n_tables) {
2492 table = &ofproto->tables[fm->table_id];
2494 return ofp_mkerr_nicira(OFPET_FLOW_MOD_FAILED, NXFMFC_BAD_TABLE_ID);
2497 /* Check for overlap, if requested. */
2498 if (fm->flags & OFPFF_CHECK_OVERLAP
2499 && classifier_rule_overlaps(table, &fm->cr)) {
2500 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP);
2503 /* Serialize against pending deletion. */
2504 if (is_flow_deletion_pending(ofproto, &fm->cr, table - ofproto->tables)) {
2505 return OFPROTO_POSTPONE;
2508 /* Allocate new rule. */
2509 rule = ofproto->ofproto_class->rule_alloc();
2511 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
2512 ofproto->name, strerror(error));
2515 rule->ofproto = ofproto;
2517 rule->pending = NULL;
2518 rule->flow_cookie = fm->cookie;
2519 rule->created = rule->modified = time_msec();
2520 rule->idle_timeout = fm->idle_timeout;
2521 rule->hard_timeout = fm->hard_timeout;
2522 rule->table_id = table - ofproto->tables;
2523 rule->send_flow_removed = (fm->flags & OFPFF_SEND_FLOW_REM) != 0;
2524 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2525 rule->n_actions = fm->n_actions;
2527 /* Insert new rule. */
2528 victim = rule_from_cls_rule(classifier_replace(table, &rule->cr));
2529 if (victim && victim->pending) {
2530 error = OFPROTO_POSTPONE;
2532 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2533 ofoperation_create(group, rule, OFOPERATION_ADD);
2534 rule->pending->victim = victim;
2536 error = ofproto->ofproto_class->rule_construct(rule);
2538 ofoperation_destroy(rule->pending);
2540 ofopgroup_submit(group);
2543 /* Back out if an error occurred. */
2546 classifier_replace(table, &victim->cr);
2548 classifier_remove(table, &rule->cr);
2550 ofproto_rule_destroy__(rule);
2555 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
2557 /* Modifies the rules listed in 'rules', changing their actions to match those
2560 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2563 * Returns 0 on success, otherwise an OpenFlow error code. */
2565 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2566 const struct ofputil_flow_mod *fm,
2567 const struct ofp_header *request, struct list *rules)
2569 struct ofopgroup *group;
2572 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
2573 LIST_FOR_EACH (rule, ofproto_node, rules) {
2574 if (!ofputil_actions_equal(fm->actions, fm->n_actions,
2575 rule->actions, rule->n_actions)) {
2576 ofoperation_create(group, rule, OFOPERATION_MODIFY);
2577 rule->pending->actions = rule->actions;
2578 rule->pending->n_actions = rule->n_actions;
2579 rule->actions = ofputil_actions_clone(fm->actions, fm->n_actions);
2580 rule->n_actions = fm->n_actions;
2581 rule->ofproto->ofproto_class->rule_modify_actions(rule);
2583 rule->modified = time_msec();
2585 rule->flow_cookie = fm->cookie;
2587 ofopgroup_submit(group);
2592 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
2593 * encoded by ofp_mkerr() on failure.
2595 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2598 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2599 const struct ofputil_flow_mod *fm,
2600 const struct ofp_header *request)
2605 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, OFPP_NONE,
2607 return (error ? error
2608 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2609 : modify_flows__(ofproto, ofconn, fm, request, &rules));
2612 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
2613 * code as encoded by ofp_mkerr() on failure.
2615 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
2618 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2619 const struct ofputil_flow_mod *fm,
2620 const struct ofp_header *request)
2625 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, OFPP_NONE,
2627 return (error ? error
2628 : list_is_empty(&rules) ? add_flow(ofproto, ofconn, fm, request)
2629 : list_is_singleton(&rules) ? modify_flows__(ofproto, ofconn,
2630 fm, request, &rules)
2634 /* OFPFC_DELETE implementation. */
2636 /* Deletes the rules listed in 'rules'.
2638 * Returns 0 on success, otherwise an OpenFlow error code. */
2640 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
2641 const struct ofp_header *request, struct list *rules)
2643 struct rule *rule, *next;
2644 struct ofopgroup *group;
2646 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
2647 LIST_FOR_EACH_SAFE (rule, next, ofproto_node, rules) {
2648 ofproto_rule_send_removed(rule, OFPRR_DELETE);
2650 ofoperation_create(group, rule, OFOPERATION_DELETE);
2651 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2652 rule->ofproto->ofproto_class->rule_destruct(rule);
2654 ofopgroup_submit(group);
2659 /* Implements OFPFC_DELETE. */
2661 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
2662 const struct ofputil_flow_mod *fm,
2663 const struct ofp_header *request)
2668 error = collect_rules_loose(ofproto, fm->table_id, &fm->cr, fm->out_port,
2670 return (error ? error
2671 : !list_is_empty(&rules) ? delete_flows__(ofproto, ofconn, request,
2676 /* Implements OFPFC_DELETE_STRICT. */
2678 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
2679 const struct ofputil_flow_mod *fm,
2680 const struct ofp_header *request)
2685 error = collect_rules_strict(ofproto, fm->table_id, &fm->cr, fm->out_port,
2687 return (error ? error
2688 : list_is_singleton(&rules) ? delete_flows__(ofproto, ofconn,
2694 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
2696 struct ofputil_flow_removed fr;
2698 if (rule_is_hidden(rule) || !rule->send_flow_removed) {
2703 fr.cookie = rule->flow_cookie;
2705 calc_flow_duration__(rule->created, &fr.duration_sec, &fr.duration_nsec);
2706 fr.idle_timeout = rule->idle_timeout;
2707 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
2710 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
2713 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
2714 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
2717 * ofproto implementation ->run() functions should use this function to expire
2718 * OpenFlow flows. */
2720 ofproto_rule_expire(struct rule *rule, uint8_t reason)
2722 struct ofproto *ofproto = rule->ofproto;
2723 struct ofopgroup *group;
2725 assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT);
2727 ofproto_rule_send_removed(rule, reason);
2729 group = ofopgroup_create_unattached(ofproto);
2730 ofoperation_create(group, rule, OFOPERATION_DELETE);
2731 classifier_remove(&ofproto->tables[rule->table_id], &rule->cr);
2732 rule->ofproto->ofproto_class->rule_destruct(rule);
2733 ofopgroup_submit(group);
2737 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2739 struct ofputil_flow_mod fm;
2742 error = reject_slave_controller(ofconn);
2747 error = ofputil_decode_flow_mod(&fm, oh,
2748 ofconn_get_flow_mod_table_id(ofconn));
2753 /* We do not support the emergency flow cache. It will hopefully get
2754 * dropped from OpenFlow in the near future. */
2755 if (fm.flags & OFPFF_EMERG) {
2756 /* There isn't a good fit for an error code, so just state that the
2757 * flow table is full. */
2758 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL);
2761 return handle_flow_mod__(ofconn_get_ofproto(ofconn), ofconn, &fm, oh);
2765 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
2766 const struct ofputil_flow_mod *fm,
2767 const struct ofp_header *oh)
2769 if (ofproto->n_pending >= 50) {
2770 assert(!list_is_empty(&ofproto->pending));
2771 return OFPROTO_POSTPONE;
2774 switch (fm->command) {
2776 return add_flow(ofproto, ofconn, fm, oh);
2779 return modify_flows_loose(ofproto, ofconn, fm, oh);
2781 case OFPFC_MODIFY_STRICT:
2782 return modify_flow_strict(ofproto, ofconn, fm, oh);
2785 return delete_flows_loose(ofproto, ofconn, fm, oh);
2787 case OFPFC_DELETE_STRICT:
2788 return delete_flow_strict(ofproto, ofconn, fm, oh);
2791 if (fm->command > 0xff) {
2792 VLOG_WARN_RL(&rl, "flow_mod has explicit table_id but "
2793 "flow_mod_table_id extension is not enabled");
2795 return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND);
2800 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
2802 struct nx_role_request *nrr = (struct nx_role_request *) oh;
2803 struct nx_role_request *reply;
2807 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY) {
2808 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2811 role = ntohl(nrr->role);
2812 if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER
2813 && role != NX_ROLE_SLAVE) {
2814 return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_ROLE);
2817 if (ofconn_get_role(ofconn) != role
2818 && ofconn_has_pending_opgroups(ofconn)) {
2819 return OFPROTO_POSTPONE;
2822 ofconn_set_role(ofconn, role);
2824 reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, oh->xid, &buf);
2825 reply->role = htonl(role);
2826 ofconn_send_reply(ofconn, buf);
2832 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
2833 const struct ofp_header *oh)
2835 const struct nxt_flow_mod_table_id *msg
2836 = (const struct nxt_flow_mod_table_id *) oh;
2838 ofconn_set_flow_mod_table_id(ofconn, msg->set != 0);
2843 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
2845 const struct nxt_set_flow_format *msg
2846 = (const struct nxt_set_flow_format *) oh;
2849 format = ntohl(msg->format);
2850 if (format != NXFF_OPENFLOW10 && format != NXFF_NXM) {
2851 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM);
2854 if (format != ofconn_get_flow_format(ofconn)
2855 && ofconn_has_pending_opgroups(ofconn)) {
2856 /* Avoid sending async messages in surprising flow format. */
2857 return OFPROTO_POSTPONE;
2860 ofconn_set_flow_format(ofconn, format);
2865 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
2867 struct ofp_header *ob;
2870 if (ofconn_has_pending_opgroups(ofconn)) {
2871 return OFPROTO_POSTPONE;
2874 ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf);
2875 ofconn_send_reply(ofconn, buf);
2880 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
2882 const struct ofp_header *oh = msg->data;
2883 const struct ofputil_msg_type *type;
2886 error = ofputil_decode_msg_type(oh, &type);
2891 switch (ofputil_msg_type_code(type)) {
2892 /* OpenFlow requests. */
2893 case OFPUTIL_OFPT_ECHO_REQUEST:
2894 return handle_echo_request(ofconn, oh);
2896 case OFPUTIL_OFPT_FEATURES_REQUEST:
2897 return handle_features_request(ofconn, oh);
2899 case OFPUTIL_OFPT_GET_CONFIG_REQUEST:
2900 return handle_get_config_request(ofconn, oh);
2902 case OFPUTIL_OFPT_SET_CONFIG:
2903 return handle_set_config(ofconn, msg->data);
2905 case OFPUTIL_OFPT_PACKET_OUT:
2906 return handle_packet_out(ofconn, oh);
2908 case OFPUTIL_OFPT_PORT_MOD:
2909 return handle_port_mod(ofconn, oh);
2911 case OFPUTIL_OFPT_FLOW_MOD:
2912 return handle_flow_mod(ofconn, oh);
2914 case OFPUTIL_OFPT_BARRIER_REQUEST:
2915 return handle_barrier_request(ofconn, oh);
2917 /* OpenFlow replies. */
2918 case OFPUTIL_OFPT_ECHO_REPLY:
2921 /* Nicira extension requests. */
2922 case OFPUTIL_NXT_ROLE_REQUEST:
2923 return handle_role_request(ofconn, oh);
2925 case OFPUTIL_NXT_FLOW_MOD_TABLE_ID:
2926 return handle_nxt_flow_mod_table_id(ofconn, oh);
2928 case OFPUTIL_NXT_SET_FLOW_FORMAT:
2929 return handle_nxt_set_flow_format(ofconn, oh);
2931 case OFPUTIL_NXT_FLOW_MOD:
2932 return handle_flow_mod(ofconn, oh);
2934 /* Statistics requests. */
2935 case OFPUTIL_OFPST_DESC_REQUEST:
2936 return handle_desc_stats_request(ofconn, msg->data);
2938 case OFPUTIL_OFPST_FLOW_REQUEST:
2939 case OFPUTIL_NXST_FLOW_REQUEST:
2940 return handle_flow_stats_request(ofconn, msg->data);
2942 case OFPUTIL_OFPST_AGGREGATE_REQUEST:
2943 case OFPUTIL_NXST_AGGREGATE_REQUEST:
2944 return handle_aggregate_stats_request(ofconn, msg->data);
2946 case OFPUTIL_OFPST_TABLE_REQUEST:
2947 return handle_table_stats_request(ofconn, msg->data);
2949 case OFPUTIL_OFPST_PORT_REQUEST:
2950 return handle_port_stats_request(ofconn, msg->data);
2952 case OFPUTIL_OFPST_QUEUE_REQUEST:
2953 return handle_queue_stats_request(ofconn, msg->data);
2955 case OFPUTIL_MSG_INVALID:
2956 case OFPUTIL_OFPT_HELLO:
2957 case OFPUTIL_OFPT_ERROR:
2958 case OFPUTIL_OFPT_FEATURES_REPLY:
2959 case OFPUTIL_OFPT_GET_CONFIG_REPLY:
2960 case OFPUTIL_OFPT_PACKET_IN:
2961 case OFPUTIL_OFPT_FLOW_REMOVED:
2962 case OFPUTIL_OFPT_PORT_STATUS:
2963 case OFPUTIL_OFPT_BARRIER_REPLY:
2964 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST:
2965 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY:
2966 case OFPUTIL_OFPST_DESC_REPLY:
2967 case OFPUTIL_OFPST_FLOW_REPLY:
2968 case OFPUTIL_OFPST_QUEUE_REPLY:
2969 case OFPUTIL_OFPST_PORT_REPLY:
2970 case OFPUTIL_OFPST_TABLE_REPLY:
2971 case OFPUTIL_OFPST_AGGREGATE_REPLY:
2972 case OFPUTIL_NXT_ROLE_REPLY:
2973 case OFPUTIL_NXT_FLOW_REMOVED:
2974 case OFPUTIL_NXST_FLOW_REPLY:
2975 case OFPUTIL_NXST_AGGREGATE_REPLY:
2977 if (oh->type == OFPT_STATS_REQUEST || oh->type == OFPT_STATS_REPLY) {
2978 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT);
2980 return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE);
2986 handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg)
2988 int error = handle_openflow__(ofconn, ofp_msg);
2989 if (error && error != OFPROTO_POSTPONE) {
2990 ofconn_send_error(ofconn, ofp_msg->data, error);
2992 COVERAGE_INC(ofproto_recv_openflow);
2993 return error != OFPROTO_POSTPONE;
2996 /* Asynchronous operations. */
2998 /* Creates and returns a new ofopgroup that is not associated with any
2999 * OpenFlow connection.
3001 * The caller should add operations to the returned group with
3002 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3003 static struct ofopgroup *
3004 ofopgroup_create_unattached(struct ofproto *ofproto)
3006 struct ofopgroup *group = xzalloc(sizeof *group);
3007 group->ofproto = ofproto;
3008 list_init(&group->ofproto_node);
3009 list_init(&group->ops);
3010 list_init(&group->ofconn_node);
3014 /* Creates and returns a new ofopgroup for 'ofproto'.
3016 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
3017 * connection. The 'request' and 'buffer_id' arguments are ignored.
3019 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
3020 * If the ofopgroup eventually fails, then the error reply will include
3021 * 'request'. If the ofopgroup eventually succeeds, then the packet with
3022 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
3024 * The caller should add operations to the returned group with
3025 * ofoperation_create() and then submit it with ofopgroup_submit(). */
3026 static struct ofopgroup *
3027 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
3028 const struct ofp_header *request, uint32_t buffer_id)
3030 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3032 size_t request_len = ntohs(request->length);
3034 assert(ofconn_get_ofproto(ofconn) == ofproto);
3036 ofconn_add_opgroup(ofconn, &group->ofconn_node);
3037 group->ofconn = ofconn;
3038 group->request = xmemdup(request, MIN(request_len, 64));
3039 group->buffer_id = buffer_id;
3044 /* Submits 'group' for processing.
3046 * If 'group' contains no operations (e.g. none were ever added, or all of the
3047 * ones that were added completed synchronously), then it is destroyed
3048 * immediately. Otherwise it is added to the ofproto's list of pending
3051 ofopgroup_submit(struct ofopgroup *group)
3053 if (list_is_empty(&group->ops)) {
3054 ofopgroup_destroy(group);
3056 list_push_back(&group->ofproto->pending, &group->ofproto_node);
3057 group->ofproto->n_pending++;
3062 ofopgroup_destroy(struct ofopgroup *group)
3064 assert(list_is_empty(&group->ops));
3065 if (!list_is_empty(&group->ofproto_node)) {
3066 assert(group->ofproto->n_pending > 0);
3067 group->ofproto->n_pending--;
3068 list_remove(&group->ofproto_node);
3070 if (!list_is_empty(&group->ofconn_node)) {
3071 list_remove(&group->ofconn_node);
3073 ofconn_send_error(group->ofconn, group->request, group->error);
3075 connmgr_retry(group->ofproto->connmgr);
3077 free(group->request);
3081 /* Initiates a new operation on 'rule', of the specified 'type', within
3082 * 'group'. Prior to calling, 'rule' must not have any pending operation. */
3084 ofoperation_create(struct ofopgroup *group, struct rule *rule,
3085 enum ofoperation_type type)
3087 struct ofoperation *op;
3089 assert(!rule->pending);
3091 op = rule->pending = xzalloc(sizeof *op);
3093 list_push_back(&group->ops, &op->group_node);
3097 op->flow_cookie = rule->flow_cookie;
3099 if (type == OFOPERATION_DELETE) {
3100 hmap_insert(&op->group->ofproto->deletions, &op->hmap_node,
3101 cls_rule_hash(&rule->cr, rule->table_id));
3106 ofoperation_destroy(struct ofoperation *op)
3108 struct ofopgroup *group = op->group;
3111 op->rule->pending = NULL;
3113 if (op->type == OFOPERATION_DELETE) {
3114 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
3116 list_remove(&op->group_node);
3120 if (list_is_empty(&group->ops) && !list_is_empty(&group->ofproto_node)) {
3121 ofopgroup_destroy(group);
3125 /* Indicates that 'op' completed with status 'error', which is either 0 to
3126 * indicate success or an OpenFlow error code (constructed with
3127 * e.g. ofp_mkerr()).
3129 * If 'error' is 0, indicating success, the operation will be committed
3130 * permanently to the flow table. There is one interesting subcase:
3132 * - If 'op' is an "add flow" operation that is replacing an existing rule in
3133 * the flow table (the "victim" rule) by a new one, then the caller must
3134 * have uninitialized any derived state in the victim rule, as in step 5 in
3135 * the "Life Cycle" in ofproto/ofproto-provider.h. ofoperation_complete()
3136 * performs steps 6 and 7 for the victim rule, most notably by calling its
3137 * ->rule_dealloc() function.
3139 * If 'error' is nonzero, then generally the operation will be rolled back:
3141 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
3142 * restores the original rule. The caller must have uninitialized any
3143 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
3144 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
3145 * and 7 for the new rule, calling its ->rule_dealloc() function.
3147 * - If 'op' is a "modify flow" operation, ofproto restores the original
3150 * - 'op' must not be a "delete flow" operation. Removing a rule is not
3151 * allowed to fail. It must always succeed.
3153 * Please see the large comment in ofproto/ofproto-provider.h titled
3154 * "Asynchronous Operation Support" for more information. */
3156 ofoperation_complete(struct ofoperation *op, int error)
3158 struct ofopgroup *group = op->group;
3159 struct rule *rule = op->rule;
3160 struct ofproto *ofproto = rule->ofproto;
3161 struct classifier *table = &ofproto->tables[rule->table_id];
3163 assert(rule->pending == op);
3164 assert(op->status < 0);
3169 && op->type != OFOPERATION_DELETE
3171 && group->buffer_id != UINT32_MAX
3172 && list_is_singleton(&op->group_node)) {
3173 struct ofpbuf *packet;
3176 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
3180 error = rule_execute(rule, in_port, packet);
3183 if (!group->error) {
3184 group->error = error;
3188 case OFOPERATION_ADD:
3191 ofproto_rule_destroy__(op->victim);
3193 if ((rule->cr.wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3194 == htons(VLAN_VID_MASK)) {
3195 if (ofproto->vlan_bitmap) {
3196 uint16_t vid = vlan_tci_to_vid(rule->cr.flow.vlan_tci);
3198 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
3199 bitmap_set1(ofproto->vlan_bitmap, vid);
3200 ofproto->vlans_changed = true;
3203 ofproto->vlans_changed = true;
3208 classifier_replace(table, &op->victim->cr);
3211 classifier_remove(table, &rule->cr);
3213 ofproto_rule_destroy__(rule);
3218 case OFOPERATION_DELETE:
3220 ofproto_rule_destroy__(rule);
3224 case OFOPERATION_MODIFY:
3226 rule->modified = time_msec();
3228 free(rule->actions);
3229 rule->actions = op->actions;
3230 rule->n_actions = op->n_actions;
3238 ofoperation_destroy(op);
3242 ofoperation_get_victim(struct ofoperation *op)
3244 assert(op->type == OFOPERATION_ADD);
3249 pick_datapath_id(const struct ofproto *ofproto)
3251 const struct ofport *port;
3253 port = ofproto_get_port(ofproto, OFPP_LOCAL);
3255 uint8_t ea[ETH_ADDR_LEN];
3258 error = netdev_get_etheraddr(port->netdev, ea);
3260 return eth_addr_to_uint64(ea);
3262 VLOG_WARN("could not get MAC address for %s (%s)",
3263 netdev_get_name(port->netdev), strerror(error));
3265 return ofproto->fallback_dpid;
3269 pick_fallback_dpid(void)
3271 uint8_t ea[ETH_ADDR_LEN];
3272 eth_addr_nicira_random(ea);
3273 return eth_addr_to_uint64(ea);
3276 /* unixctl commands. */
3279 ofproto_lookup(const char *name)
3281 struct ofproto *ofproto;
3283 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
3285 if (!strcmp(ofproto->name, name)) {
3293 ofproto_unixctl_list(struct unixctl_conn *conn, const char *arg OVS_UNUSED,
3294 void *aux OVS_UNUSED)
3296 struct ofproto *ofproto;
3300 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
3301 ds_put_format(&results, "%s\n", ofproto->name);
3303 unixctl_command_reply(conn, 200, ds_cstr(&results));
3304 ds_destroy(&results);
3308 ofproto_unixctl_init(void)
3310 static bool registered;
3316 unixctl_command_register("ofproto/list", "", ofproto_unixctl_list, NULL);
3319 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
3321 * This is deprecated. It is only for compatibility with broken device drivers
3322 * in old versions of Linux that do not properly support VLANs when VLAN
3323 * devices are not used. When broken device drivers are no longer in
3324 * widespread use, we will delete these interfaces. */
3326 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
3327 * (exactly) by an OpenFlow rule in 'ofproto'. */
3329 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
3331 const struct classifier *cls;
3333 free(ofproto->vlan_bitmap);
3334 ofproto->vlan_bitmap = bitmap_allocate(4096);
3335 ofproto->vlans_changed = false;
3337 OFPROTO_FOR_EACH_TABLE (cls, ofproto) {
3338 const struct cls_table *table;
3340 HMAP_FOR_EACH (table, hmap_node, &cls->tables) {
3341 if ((table->wc.vlan_tci_mask & htons(VLAN_VID_MASK))
3342 == htons(VLAN_VID_MASK)) {
3343 const struct cls_rule *rule;
3345 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
3346 uint16_t vid = vlan_tci_to_vid(rule->flow.vlan_tci);
3347 bitmap_set1(vlan_bitmap, vid);
3348 bitmap_set1(ofproto->vlan_bitmap, vid);
3355 /* Returns true if new VLANs have come into use by the flow table since the
3356 * last call to ofproto_get_vlan_usage().
3358 * We don't track when old VLANs stop being used. */
3360 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
3362 return ofproto->vlans_changed;
3365 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
3366 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
3367 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
3368 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
3369 * then the VLAN device is un-enslaved. */
3371 ofproto_port_set_realdev(struct ofproto *ofproto, uint16_t vlandev_ofp_port,
3372 uint16_t realdev_ofp_port, int vid)
3374 struct ofport *ofport;
3377 assert(vlandev_ofp_port != realdev_ofp_port);
3379 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
3381 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
3382 ofproto->name, vlandev_ofp_port);
3386 if (!ofproto->ofproto_class->set_realdev) {
3387 if (!vlandev_ofp_port) {
3390 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
3394 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
3396 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
3397 ofproto->name, vlandev_ofp_port,
3398 netdev_get_name(ofport->netdev), strerror(error));